Skip to content

P
Paddle

橘蓝 / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle

体验新版 GitCode，发现更多精彩内容 >>

提交 84cab2c7 编写于 作者: L liaogang
浏览文件
操作

Merge conflict with develop branch

上级 978a6e5e 9bce3289
展开全部 隐藏空白更改
内联 并排
Showing with 13576 addition and 9734 deletion
.clang-format
浏览文件 @ 84cab2c7
......@@ -13,8 +13,6 @@
# The document of clang-format is
# http://clang.llvm.org/docs/ClangFormat.html
# http://clang.llvm.org/docs/ClangFormatStyleOptions.html
#
# TODO(yuyang18): Add python and other language code style
---
Language: Cpp
BasedOnStyle: Google
......@@ -22,8 +20,9 @@ IndentWidth: 2
TabWidth: 2
ContinuationIndentWidth: 4
AccessModifierOffset: -2 # The private/protected/public has no indent in class
PointerAlignment: Left # int* p/int& p, not int *p/int &p
Standard: Cpp11
AllowAllParametersOfDeclarationOnNextLine: true
BinPackParameters: false
BinPackArguments: false
...
.gitignore
浏览文件 @ 84cab2c7
......@@ -5,4 +5,6 @@ build/
.vscode
.idea
.project
.cproject
.pydevproject
Makefile
.pre-commit-config.yaml 0 → 100644
浏览文件 @ 84cab2c7
- repo: https://github.com/Lucas-C/pre-commit-hooks.git
sha: c25201a00e6b0514370501050cf2a8538ac12270
hooks:
- id: remove-crlf
- repo: https://github.com/reyoung/mirrors-yapf.git
sha: v0.13.2
hooks:
- id: yapf
- repo: https://github.com/pre-commit/pre-commit-hooks
sha: 4ef03c4223ad322c7adaa6c6c0efb26b57df3b71
hooks:
- id: check-added-large-files
- id: check-merge-conflict
- id: check-symlinks
- id: detect-private-key
- id: end-of-file-fixer
# TODO(yuyang): trailing whitespace has some bugs on markdown
# files now, please not add it to pre-commit hook now
# - id: trailing-whitespace
#
# TODO(yuyang): debug-statements not fit for Paddle, because
# not all of our python code is runnable. Some are used for
# documenation
# - id: debug-statements
.style.yapf 0 → 100644
浏览文件 @ 84cab2c7
[style]
based_on_style = pep8
column_limit = 80
demo/image_classification/data/process_cifar.py
浏览文件 @ 84cab2c7
......@@ -16,7 +16,6 @@ import numpy as np
import sys
import os
import PIL.Image as Image
"""
Usage: python process_cifar input_dir output_dir
"""
......@@ -30,6 +29,7 @@ def mkdir_not_exist(path):
if not os.path.exists(path):
os.mkdir(path)
def create_dir_structure(output_dir):
"""
Create the directory structure for the directory.
......@@ -39,8 +39,8 @@ def create_dir_structure(output_dir):
mkdir_not_exist(os.path.join(output_dir, "train"))
mkdir_not_exist(os.path.join(output_dir, "test"))
def convert_batch(batch_path, label_set, label_map,
output_dir, data_split):
def convert_batch(batch_path, label_set, label_map, output_dir, data_split):
"""
Convert CIFAR batch to the structure of Paddle format.
batch_path: the batch to be converted.
......@@ -67,11 +67,23 @@ if __name__ == '__main__':
output_dir = sys.argv[2]
num_batch = 5
create_dir_structure(output_dir)
label_map = {0: "airplane", 1: "automobile", 2: "bird", 3: "cat", 4: "deer",
5: "dog", 6: "frog", 7: "horse", 8: "ship", 9: "truck"}
label_map = {
0: "airplane",
1: "automobile",
2: "bird",
3: "cat",
4: "deer",
5: "dog",
6: "frog",
7: "horse",
8: "ship",
9: "truck"
}
labels = {}
for i in range(1, num_batch + 1):
convert_batch(os.path.join(input_dir, "data_batch_%d" % i), labels,
label_map, output_dir, "train")
convert_batch(os.path.join(input_dir, "test_batch"), {},
label_map, output_dir, "test")
\ No newline at end of file
convert_batch(
os.path.join(input_dir, "data_batch_%d" % i), labels, label_map,
output_dir, "train")
convert_batch(
os.path.join(input_dir, "test_batch"), {}, label_map, output_dir,
"test")
demo/image_classification/image_provider.py
浏览文件 @ 84cab2c7
......@@ -46,14 +46,14 @@ def hook(settings, img_size, mean_img_size, num_classes, color, meta, use_jpeg,
settings.img_mean = image_util.load_meta(settings.meta_path,
settings.mean_img_size,
settings.img_size,
settings.color)
settings.img_size, settings.color)
settings.logger.info('Image size: %s', settings.img_size)
settings.logger.info('Meta path: %s', settings.meta_path)
settings.input_types = [
dense_vector(settings.img_raw_size), # image feature
integer_value(settings.num_classes)] # labels
integer_value(settings.num_classes)
] # labels
settings.logger.info('DataProvider Initialization finished')
......@@ -79,8 +79,8 @@ def processData(settings, file_list):
img = image_util.decode_jpeg(data['images'][i])
else:
img = data['images'][i]
img_feat = image_util.preprocess_img(img, settings.img_mean,
settings.img_size, settings.is_train,
settings.color)
img_feat = image_util.preprocess_img(
img, settings.img_mean, settings.img_size,
settings.is_train, settings.color)
label = data['labels'][i]
yield img_feat.astype('float32'), int(label)
demo/image_classification/image_util.py
浏览文件 @ 84cab2c7
......@@ -16,17 +16,20 @@ import numpy as np
from PIL import Image
from cStringIO import StringIO
def resize_image(img, target_size):
"""
Resize an image so that the shorter edge has length target_size.
img: the input image to be resized.
target_size: the target resized image size.
"""
percent = (target_size/float(min(img.size[0], img.size[1])))
resized_size = int(round(img.size[0] * percent)), int(round(img.size[1] * percent))
percent = (target_size / float(min(img.size[0], img.size[1])))
resized_size = int(round(img.size[0] * percent)), int(
round(img.size[1] * percent))
img = img.resize(resized_size, Image.ANTIALIAS)
return img
def flip(im):
"""
Return the flipped image.
......@@ -38,6 +41,7 @@ def flip(im):
else:
return im[:, ::-1]
def crop_img(im, inner_size, color=True, test=True):
"""
Return cropped image.
......@@ -50,20 +54,22 @@ def crop_img(im, inner_size, color=True, test=True):
If True, crop the center of images.
"""
if color:
height, width = max(inner_size, im.shape[1]), max(inner_size, im.shape[2])
height, width = max(inner_size, im.shape[1]), max(inner_size,
im.shape[2])
padded_im = np.zeros((3, height, width))
startY = (height - im.shape[1]) / 2
startX = (width - im.shape[2]) / 2
endY, endX = startY + im.shape[1], startX + im.shape[2]
padded_im[:, startY: endY, startX: endX] = im
padded_im[:, startY:endY, startX:endX] = im
else:
im = im.astype('float32')
height, width = max(inner_size, im.shape[0]), max(inner_size, im.shape[1])
height, width = max(inner_size, im.shape[0]), max(inner_size,
im.shape[1])
padded_im = np.zeros((height, width))
startY = (height - im.shape[0]) / 2
startX = (width - im.shape[1]) / 2
endY, endX = startY + im.shape[0], startX + im.shape[1]
padded_im[startY: endY, startX: endX] = im
padded_im[startY:endY, startX:endX] = im
if test:
startY = (height - inner_size) / 2
startX = (width - inner_size) / 2
......@@ -72,19 +78,21 @@ def crop_img(im, inner_size, color=True, test=True):
startX = np.random.randint(0, width - inner_size + 1)
endY, endX = startY + inner_size, startX + inner_size
if color:
pic = padded_im[:, startY: endY, startX: endX]
pic = padded_im[:, startY:endY, startX:endX]
else:
pic = padded_im[startY: endY, startX: endX]
pic = padded_im[startY:endY, startX:endX]
if (not test) and (np.random.randint(2) == 0):
pic = flip(pic)
return pic
def decode_jpeg(jpeg_string):
np_array = np.array(Image.open(StringIO(jpeg_string)))
if len(np_array.shape) == 3:
np_array = np.transpose(np_array, (2, 0, 1))
return np_array
def preprocess_img(im, img_mean, crop_size, is_train, color=True):
"""
Does data augmentation for images.
......@@ -99,6 +107,7 @@ def preprocess_img(im, img_mean, crop_size, is_train, color=True):
pic -= img_mean
return pic.flatten()
def load_meta(meta_path, mean_img_size, crop_size, color=True):
"""
Return the loaded meta file.
......@@ -109,17 +118,18 @@ def load_meta(meta_path, mean_img_size, crop_size, color=True):
mean = np.load(meta_path)['data_mean']
border = (mean_img_size - crop_size) / 2
if color:
assert(mean_img_size * mean_img_size * 3 == mean.shape[0])
assert (mean_img_size * mean_img_size * 3 == mean.shape[0])
mean = mean.reshape(3, mean_img_size, mean_img_size)
mean = mean[:, border: border + crop_size,
border: border + crop_size].astype('float32')
mean = mean[:, border:border + crop_size, border:border +
crop_size].astype('float32')
else:
assert(mean_img_size * mean_img_size == mean.shape[0])
assert (mean_img_size * mean_img_size == mean.shape[0])
mean = mean.reshape(mean_img_size, mean_img_size)
mean = mean[border: border + crop_size,
border: border + crop_size].astype('float32')
mean = mean[border:border + crop_size, border:border +
crop_size].astype('float32')
return mean
def load_image(img_path, is_color=True):
"""
Load image and return.
......@@ -130,6 +140,7 @@ def load_image(img_path, is_color=True):
img.load()
return img
def oversample(img, crop_dims):
"""
image : iterable of (H x W x K) ndarrays
......@@ -152,50 +163,53 @@ def oversample(img, crop_dims):
for j in w_indices:
crops_ix[curr] = (i, j, i + crop_dims[0], j + crop_dims[1])
curr += 1
crops_ix[4] = np.tile(im_center, (1, 2)) + np.concatenate([
-crop_dims / 2.0,
crop_dims / 2.0
])
crops_ix[4] = np.tile(im_center, (1, 2)) + np.concatenate(
[-crop_dims / 2.0, crop_dims / 2.0])
crops_ix = np.tile(crops_ix, (2, 1))
# Extract crops
crops = np.empty((10 * len(img), crop_dims[0], crop_dims[1],
im_shape[-1]), dtype=np.float32)
crops = np.empty(
(10 * len(img), crop_dims[0], crop_dims[1], im_shape[-1]),
dtype=np.float32)
ix = 0
for im in img:
for crop in crops_ix:
crops[ix] = im[crop[0]:crop[2], crop[1]:crop[3], :]
ix += 1
crops[ix-5:ix] = crops[ix-5:ix, :, ::-1, :] # flip for mirrors
crops[ix - 5:ix] = crops[ix - 5:ix, :, ::-1, :] # flip for mirrors
return crops
class ImageTransformer:
def __init__(self, transpose = None,
channel_swap = None, mean = None, is_color = True):
def __init__(self,
transpose=None,
channel_swap=None,
mean=None,
is_color=True):
self.transpose = transpose
self.channel_swap = None
self.mean = None
self.is_color = is_color
self.is_color = is_color
def set_transpose(self, order):
def set_transpose(self, order):
if self.is_color:
assert 3 == len(order)
assert 3 == len(order)
self.transpose = order
def set_channel_swap(self, order):
def set_channel_swap(self, order):
if self.is_color:
assert 3 == len(order)
assert 3 == len(order)
self.channel_swap = order
def set_mean(self, mean):
# mean value, may be one value per channel
if mean.ndim == 1:
mean = mean[:, np.newaxis, np.newaxis]
else:
mean = mean[:, np.newaxis, np.newaxis]
else:
# elementwise mean
if self.is_color:
assert len(mean.shape) == 3
self.mean = mean
self.mean = mean
def transformer(self, data):
if self.transpose is not None:
......
demo/image_classification/prediction.py
浏览文件 @ 84cab2c7
......@@ -12,7 +12,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
import os,sys
import os, sys
import numpy as np
import logging
from PIL import Image
......@@ -24,9 +24,11 @@ from py_paddle import swig_paddle, DataProviderConverter
from paddle.trainer.PyDataProvider2 import dense_vector
from paddle.trainer.config_parser import parse_config
logging.basicConfig(format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s')
logging.basicConfig(
format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s')
logging.getLogger().setLevel(logging.INFO)
class ImageClassifier():
def __init__(self,
train_conf,
......@@ -58,18 +60,19 @@ class ImageClassifier():
self.oversample = oversample
self.is_color = is_color
self.transformer = image_util.ImageTransformer(is_color = is_color)
self.transformer.set_transpose((2,0,1))
self.transformer = image_util.ImageTransformer(is_color=is_color)
self.transformer.set_transpose((2, 0, 1))
self.mean_file = mean_file
mean = np.load(self.mean_file)['data_mean']
mean = mean.reshape(3, self.crop_dims[0], self.crop_dims[1])
self.transformer.set_mean(mean) # mean pixel
self.transformer.set_mean(mean) # mean pixel
gpu = 1 if use_gpu else 0
conf_args = "is_test=1,use_gpu=%d,is_predict=1" % (gpu)
conf = parse_config(train_conf, conf_args)
swig_paddle.initPaddle("--use_gpu=%d" % (gpu))
self.network = swig_paddle.GradientMachine.createFromConfigProto(conf.model_config)
self.network = swig_paddle.GradientMachine.createFromConfigProto(
conf.model_config)
assert isinstance(self.network, swig_paddle.GradientMachine)
self.network.loadParameters(self.model_dir)
......@@ -90,14 +93,14 @@ class ImageClassifier():
# image_util.resize_image: short side is self.resize_dim
image = image_util.resize_image(image, self.resize_dim)
image = np.array(image)
input = np.zeros((1, image.shape[0], image.shape[1], 3),
dtype=np.float32)
input = np.zeros(
(1, image.shape[0], image.shape[1], 3), dtype=np.float32)
input[0] = image.astype(np.float32)
input = image_util.oversample(input, self.crop_dims)
else:
image = image.resize(self.crop_dims, Image.ANTIALIAS)
input = np.zeros((1, self.crop_dims[0], self.crop_dims[1], 3),
dtype=np.float32)
input = np.zeros(
(1, self.crop_dims[0], self.crop_dims[1], 3), dtype=np.float32)
input[0] = np.array(image).astype(np.float32)
data_in = []
......@@ -133,22 +136,24 @@ class ImageClassifier():
lab = np.argsort(-prob)
logging.info("Label of %s is: %d", image, lab[0])
if __name__ == '__main__':
image_size=32
crop_size=32
multi_crop=True
config="vgg_16_cifar.py"
output_layer="__fc_layer_1__"
mean_path="data/cifar-out/batches/batches.meta"
model_path=sys.argv[1]
image=sys.argv[2]
use_gpu=bool(int(sys.argv[3]))
obj = ImageClassifier(train_conf=config,
model_dir=model_path,
resize_dim=image_size,
crop_dim=crop_size,
mean_file=mean_path,
use_gpu=use_gpu,
oversample=multi_crop)
image_size = 32
crop_size = 32
multi_crop = True
config = "vgg_16_cifar.py"
output_layer = "__fc_layer_1__"
mean_path = "data/cifar-out/batches/batches.meta"
model_path = sys.argv[1]
image = sys.argv[2]
use_gpu = bool(int(sys.argv[3]))
obj = ImageClassifier(
train_conf=config,
model_dir=model_path,
resize_dim=image_size,
crop_dim=crop_size,
mean_file=mean_path,
use_gpu=use_gpu,
oversample=multi_crop)
obj.predict(image, output_layer)
demo/image_classification/preprocess.py
浏览文件 @ 84cab2c7
......@@ -19,24 +19,36 @@ from optparse import OptionParser
def option_parser():
parser = OptionParser(usage="usage: python preprcoess.py "\
"-i data_dir [options]")
parser.add_option("-i", "--input", action="store",
dest="input", help="Input data directory.")
parser.add_option("-s", "--size", action="store",
dest="size", help="Processed image size.")
parser.add_option("-c", "--color", action="store",
dest="color", help="whether to use color images.")
parser.add_option(
"-i",
"--input",
action="store",
dest="input",
help="Input data directory.")
parser.add_option(
"-s",
"--size",
action="store",
dest="size",
help="Processed image size.")
parser.add_option(
"-c",
"--color",
action="store",
dest="color",
help="whether to use color images.")
return parser.parse_args()
if __name__ == '__main__':
options, args = option_parser()
data_dir = options.input
processed_image_size = int(options.size)
color = options.color == "1"
data_creator = ImageClassificationDatasetCreater(data_dir,
processed_image_size,
color)
data_creator.train_list_name = "train.txt"
data_creator.test_list_name = "test.txt"
data_creator.num_per_batch = 1000
data_creator.overwrite = True
data_creator.create_batches()
options, args = option_parser()
data_dir = options.input
processed_image_size = int(options.size)
color = options.color == "1"
data_creator = ImageClassificationDatasetCreater(
data_dir, processed_image_size, color)
data_creator.train_list_name = "train.txt"
data_creator.test_list_name = "test.txt"
data_creator.num_per_batch = 1000
data_creator.overwrite = True
data_creator.create_batches()
demo/image_classification/vgg_16_cifar.py
浏览文件 @ 84cab2c7
......@@ -18,36 +18,38 @@ is_predict = get_config_arg("is_predict", bool, False)
####################Data Configuration ##################
if not is_predict:
data_dir='data/cifar-out/batches/'
meta_path=data_dir+'batches.meta'
args = {'meta':meta_path,'mean_img_size': 32,
'img_size': 32,'num_classes': 10,
'use_jpeg': 1,'color': "color"}
define_py_data_sources2(train_list="train.list",
test_list="train.list",
module='image_provider',
obj='processData',
args=args)
data_dir = 'data/cifar-out/batches/'
meta_path = data_dir + 'batches.meta'
args = {
'meta': meta_path,
'mean_img_size': 32,
'img_size': 32,
'num_classes': 10,
'use_jpeg': 1,
'color': "color"
}
define_py_data_sources2(
train_list="train.list",
test_list="train.list",
module='image_provider',
obj='processData',
args=args)
######################Algorithm Configuration #############
settings(
batch_size = 128,
learning_rate = 0.1 / 128.0,
learning_method = MomentumOptimizer(0.9),
regularization = L2Regularization(0.0005 * 128)
)
batch_size=128,
learning_rate=0.1 / 128.0,
learning_method=MomentumOptimizer(0.9),
regularization=L2Regularization(0.0005 * 128))
#######################Network Configuration #############
data_size=3*32*32
label_size=10
img = data_layer(name='image',
size=data_size)
data_size = 3 * 32 * 32
label_size = 10
img = data_layer(name='image', size=data_size)
# small_vgg is predefined in trainer_config_helpers.networks
predict = small_vgg(input_image=img,
num_channels=3,
num_classes=label_size)
predict = small_vgg(input_image=img, num_channels=3, num_classes=label_size)
if not is_predict:
lbl = data_layer(name="label", size=label_size)
......
demo/introduction/README.md
浏览文件 @ 84cab2c7
This folder contains scripts used in PaddlePaddle introduction.
- use `bash train.sh` to train a simple linear regression model
- use `python evaluate_model.py` to read model parameters. You can see that `w` and `b` are very close to [2, 0.3].
demo/introduction/dataprovider.py
浏览文件 @ 84cab2c7
......@@ -15,10 +15,10 @@
from paddle.trainer.PyDataProvider2 import *
import random
# define data types of input: 2 real numbers
@provider(input_types=[dense_vector(1), dense_vector(1)],use_seq=False)
@provider(input_types=[dense_vector(1), dense_vector(1)], use_seq=False)
def process(settings, input_file):
for i in xrange(2000):
x = random.random()
yield [x], [2*x+0.3]
yield [x], [2 * x + 0.3]
demo/introduction/evaluate_model.py
浏览文件 @ 84cab2c7
......@@ -23,14 +23,17 @@ Usage:
import numpy as np
import os
def load(file_name):
with open(file_name, 'rb') as f:
f.read(16) # skip header for float type.
f.read(16) # skip header for float type.
return np.fromfile(f, dtype=np.float32)
def main():
print 'w=%.6f, b=%.6f from pass 29' % (load('output/pass-00029/w'),
load('output/pass-00029/b'))
load('output/pass-00029/b'))
if __name__ == '__main__':
main()
demo/introduction/trainer_config.py
浏览文件 @ 84cab2c7
......@@ -16,9 +16,14 @@ from paddle.trainer_config_helpers import *
# 1. read data. Suppose you saved above python code as dataprovider.py
data_file = 'empty.list'
with open(data_file, 'w') as f: f.writelines(' ')
define_py_data_sources2(train_list=data_file, test_list=None,
module='dataprovider', obj='process',args={})
with open(data_file, 'w') as f:
f.writelines(' ')
define_py_data_sources2(
train_list=data_file,
test_list=None,
module='dataprovider',
obj='process',
args={})
# 2. learning algorithm
settings(batch_size=12, learning_rate=1e-3, learning_method=MomentumOptimizer())
......@@ -26,7 +31,11 @@ settings(batch_size=12, learning_rate=1e-3, learning_method=MomentumOptimizer())
# 3. Network configuration
x = data_layer(name='x', size=1)
y = data_layer(name='y', size=1)
y_predict = fc_layer(input=x, param_attr=ParamAttr(name='w'), size=1, act=LinearActivation(), bias_attr=ParamAttr(name='b'))
y_predict = fc_layer(
input=x,
param_attr=ParamAttr(name='w'),
size=1,
act=LinearActivation(),
bias_attr=ParamAttr(name='b'))
cost = regression_cost(input=y_predict, label=y)
outputs(cost)
demo/mnist/data/generate_list.py
浏览文件 @ 84cab2c7
......@@ -13,9 +13,9 @@
# limitations under the License.
o = open("./" + "train.list", "w")
o.write("./data/raw_data/train" +"\n")
o.write("./data/raw_data/train" + "\n")
o.close()
o = open("./" + "test.list", "w")
o.write("./data/raw_data/t10k" +"\n")
o.close()
\ No newline at end of file
o.write("./data/raw_data/t10k" + "\n")
o.close()
demo/mnist/data/get_mnist_data.sh
浏览文件 @ 84cab2c7
......@@ -19,4 +19,3 @@ done
cd $DIR
rm -f *.list
python generate_list.py
demo/mnist/mnist_provider.py
浏览文件 @ 84cab2c7
......@@ -2,10 +2,9 @@ from paddle.trainer.PyDataProvider2 import *
# Define a py data provider
@provider(input_types={
'pixel': dense_vector(28 * 28),
'label': integer_value(10)
})
@provider(
input_types={'pixel': dense_vector(28 * 28),
'label': integer_value(10)})
def process(settings, filename): # settings is not used currently.
imgf = filename + "-images-idx3-ubyte"
labelf = filename + "-labels-idx1-ubyte"
......
demo/mnist/vgg_16_mnist.py
浏览文件 @ 84cab2c7
......@@ -18,32 +18,29 @@ is_predict = get_config_arg("is_predict", bool, False)
####################Data Configuration ##################
if not is_predict:
data_dir='./data/'
define_py_data_sources2(train_list= data_dir + 'train.list',
test_list= data_dir + 'test.list',
module='mnist_provider',
obj='process')
data_dir = './data/'
define_py_data_sources2(
train_list=data_dir + 'train.list',
test_list=data_dir + 'test.list',
module='mnist_provider',
obj='process')
######################Algorithm Configuration #############
settings(
batch_size = 128,
learning_rate = 0.1 / 128.0,
learning_method = MomentumOptimizer(0.9),
regularization = L2Regularization(0.0005 * 128)
)
batch_size=128,
learning_rate=0.1 / 128.0,
learning_method=MomentumOptimizer(0.9),
regularization=L2Regularization(0.0005 * 128))
#######################Network Configuration #############
data_size=1*28*28
label_size=10
data_size = 1 * 28 * 28
label_size = 10
img = data_layer(name='pixel', size=data_size)
# small_vgg is predined in trainer_config_helpers.network
predict = small_vgg(input_image=img,
num_channels=1,
num_classes=label_size)
predict = small_vgg(input_image=img, num_channels=1, num_classes=label_size)
if not is_predict:
lbl = data_layer(name="label", size=label_size)
......
demo/model_zoo/embedding/extract_para.py
浏览文件 @ 84cab2c7
......@@ -12,7 +12,6 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Example:
python extract_para.py --preModel PREMODEL --preDict PREDICT \
......@@ -29,6 +28,7 @@ Options:
from optparse import OptionParser
import struct
def get_row_index(preDict, usrDict):
"""
Get the row positions for all words in user dictionary from pre-trained dictionary.
......@@ -47,7 +47,9 @@ def get_row_index(preDict, usrDict):
pos.append(index[word])
return pos
def extract_parameters_by_usrDict(preModel, preDict, usrModel, usrDict, paraDim):
def extract_parameters_by_usrDict(preModel, preDict, usrModel, usrDict,
paraDim):
"""
Extract desired parameters from a pretrained embedding model based on user dictionary
"""
......@@ -70,6 +72,7 @@ def extract_parameters_by_usrDict(preModel, preDict, usrModel, usrDict, paraDim)
print "extract parameters finish, total", len(rowIndex), "lines"
fi.close()
def main():
"""
Main entry for running paraconvert.py
......@@ -78,19 +81,33 @@ def main():
"python %prog --preModel PREMODEL --preDict PREDICT" \
" --usrModel USRMODEL --usrDict USRDICT -d DIM"
parser = OptionParser(usage)
parser.add_option("--preModel", action="store", dest="preModel",
help="the name of pretrained embedding model")
parser.add_option("--preDict", action="store", dest="preDict",
help="the name of pretrained dictionary")
parser.add_option("--usrModel", action="store", dest="usrModel",
help="the name of output usr embedding model")
parser.add_option("--usrDict", action="store", dest="usrDict",
help="the name of user specified dictionary")
parser.add_option("-d", action="store", dest="dim",
help="dimension of parameter")
parser.add_option(
"--preModel",
action="store",
dest="preModel",
help="the name of pretrained embedding model")
parser.add_option(
"--preDict",
action="store",
dest="preDict",
help="the name of pretrained dictionary")
parser.add_option(
"--usrModel",
action="store",
dest="usrModel",
help="the name of output usr embedding model")
parser.add_option(
"--usrDict",
action="store",
dest="usrDict",
help="the name of user specified dictionary")
parser.add_option(
"-d", action="store", dest="dim", help="dimension of parameter")
(options, args) = parser.parse_args()
extract_parameters_by_usrDict(options.preModel, options.preDict,
options.usrModel, options.usrDict, int(options.dim))
extract_parameters_by_usrDict(options.preModel, options.preDict,
options.usrModel, options.usrDict,
int(options.dim))
if __name__ == '__main__':
main()
demo/model_zoo/embedding/paraconvert.py
浏览文件 @ 84cab2c7
......@@ -12,7 +12,6 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Example:
python paraconvert.py --b2t -i INPUT -o OUTPUT -d DIM
......@@ -29,6 +28,7 @@ Options:
from optparse import OptionParser
import struct
def binary2text(input, output, paraDim):
"""
Convert a binary parameter file of embedding model to be a text file.
......@@ -76,12 +76,13 @@ def binary2text(input, output, paraDim):
fo.close()
print "binary2text finish, total", line, "lines"
def get_para_count(input):
"""
Compute the total number of embedding parameters in input text file.
input: the name of input text file
"""
numRows = 1
numRows = 1
paraDim = 0
with open(input) as f:
line = f.readline()
......@@ -90,6 +91,7 @@ def get_para_count(input):
numRows += 1
return numRows * paraDim
def text2binary(input, output, paddle_head=True):
"""
Convert a text parameter file of embedding model to be a binary file.
......@@ -123,6 +125,7 @@ def text2binary(input, output, paddle_head=True):
fo.close()
print "text2binary finish, total", count, "lines"
def main():
"""
Main entry for running paraconvert.py
......@@ -131,21 +134,26 @@ def main():
"python %prog --b2t -i INPUT -o OUTPUT -d DIM \n" \
"python %prog --t2b -i INPUT -o OUTPUT"
parser = OptionParser(usage)
parser.add_option("--b2t", action="store_true",
help="convert parameter file of embedding model from binary to text")
parser.add_option("--t2b", action="store_true",
help="convert parameter file of embedding model from text to binary")
parser.add_option("-i", action="store", dest="input",
help="input parameter file name")
parser.add_option("-o", action="store", dest="output",
help="output parameter file name")
parser.add_option("-d", action="store", dest="dim",
help="dimension of parameter")
parser.add_option(
"--b2t",
action="store_true",
help="convert parameter file of embedding model from binary to text")
parser.add_option(
"--t2b",
action="store_true",
help="convert parameter file of embedding model from text to binary")
parser.add_option(
"-i", action="store", dest="input", help="input parameter file name")
parser.add_option(
"-o", action="store", dest="output", help="output parameter file name")
parser.add_option(
"-d", action="store", dest="dim", help="dimension of parameter")
(options, args) = parser.parse_args()
if options.b2t:
binary2text(options.input, options.output, options.dim)
if options.t2b:
text2binary(options.input, options.output)
if __name__ == '__main__':
main()
demo/model_zoo/resnet/classify.py
浏览文件 @ 84cab2c7
......@@ -26,16 +26,22 @@ from py_paddle import swig_paddle, DataProviderConverter
from paddle.trainer.PyDataProvider2 import dense_vector
from paddle.trainer.config_parser import parse_config
logging.basicConfig(format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s')
logging.basicConfig(
format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s')
logging.getLogger().setLevel(logging.INFO)
class ImageClassifier():
def __init__(self, train_conf, model_dir=None,
resize_dim=256, crop_dim=224,
def __init__(self,
train_conf,
model_dir=None,
resize_dim=256,
crop_dim=224,
use_gpu=True,
mean_file=None,
output_layer=None,
oversample=False, is_color=True):
oversample=False,
is_color=True):
"""
train_conf: network configure.
model_dir: string, directory of model.
......@@ -62,24 +68,25 @@ class ImageClassifier():
assert isinstance(self.output_layer, basestring)
self.output_layer = self.output_layer.split(",")
self.transformer = image_util.ImageTransformer(is_color = is_color)
self.transformer.set_transpose((2,0,1))
self.transformer.set_channel_swap((2,1,0))
self.transformer = image_util.ImageTransformer(is_color=is_color)
self.transformer.set_transpose((2, 0, 1))
self.transformer.set_channel_swap((2, 1, 0))
self.mean_file = mean_file
if self.mean_file is not None:
mean = np.load(self.mean_file)['data_mean']
mean = mean.reshape(3, self.crop_dims[0], self.crop_dims[1])
self.transformer.set_mean(mean) # mean pixel
self.transformer.set_mean(mean) # mean pixel
else:
# if you use three mean value, set like:
# this three mean value is calculated from ImageNet.
self.transformer.set_mean(np.array([103.939,116.779,123.68]))
self.transformer.set_mean(np.array([103.939, 116.779, 123.68]))
conf_args = "is_test=1,use_gpu=%d,is_predict=1" % (int(use_gpu))
conf = parse_config(train_conf, conf_args)
swig_paddle.initPaddle("--use_gpu=%d" % (int(use_gpu)))
self.network = swig_paddle.GradientMachine.createFromConfigProto(conf.model_config)
self.network = swig_paddle.GradientMachine.createFromConfigProto(
conf.model_config)
assert isinstance(self.network, swig_paddle.GradientMachine)
self.network.loadParameters(self.model_dir)
......@@ -105,14 +112,14 @@ class ImageClassifier():
# image_util.resize_image: short side is self.resize_dim
image = image_util.resize_image(image, self.resize_dim)
image = np.array(image)
input = np.zeros((1, image.shape[0], image.shape[1], 3),
dtype=np.float32)
input = np.zeros(
(1, image.shape[0], image.shape[1], 3), dtype=np.float32)
input[0] = image.astype(np.float32)
input = image_util.oversample(input, self.crop_dims)
else:
image = image.resize(self.crop_dims, Image.ANTIALIAS)
input = np.zeros((1, self.crop_dims[0], self.crop_dims[1], 3),
dtype=np.float32)
input = np.zeros(
(1, self.crop_dims[0], self.crop_dims[1], 3), dtype=np.float32)
input[0] = np.array(image).astype(np.float32)
data_in = []
......@@ -172,7 +179,7 @@ class ImageClassifier():
logging.info("Label of %s is: %d", image, lab[0])
return results
def extract(self, data_file, output_dir, batch_size = 10000):
def extract(self, data_file, output_dir, batch_size=10000):
"""
extract and save features of output layers, which are
specify in Outputs() in network configure.
......@@ -197,7 +204,7 @@ class ImageClassifier():
image_feature[file_name] = feature
sample_num += 1
if sample_num == batch_size:
batch_name = os.path.join(output_dir, 'batch_%d' %(batch_num))
batch_name = os.path.join(output_dir, 'batch_%d' % (batch_num))
self.save_file(image_feature, batch_name)
logging.info('Finish batch %d', batch_num)
batch_num += 1
......@@ -206,7 +213,7 @@ class ImageClassifier():
if idx % 1000 == 0:
logging.info('%d/%d, %s', idx, len(image_files), file_name)
if sample_num > 0:
batch_name = os.path.join(output_dir, 'batch_%d' %(batch_num))
batch_name = os.path.join(output_dir, 'batch_%d' % (batch_num))
self.save_file(image_feature, batch_name)
logging.info('Finish batch %d', batch_num)
logging.info('Done: make image feature batch')
......@@ -215,38 +222,64 @@ class ImageClassifier():
of = open(file, 'wb')
cPickle.dump(data, of, protocol=cPickle.HIGHEST_PROTOCOL)
def option_parser():
"""
Main entry for predciting
"""
usage = "%prog -c config -i data_list -w model_dir [options]"
parser = OptionParser(usage="usage: %s" % usage)
parser.add_option("-j", "--job",
action="store", dest="job_type",
help="job type: predict, extract\
parser.add_option(
"-j",
"--job",
action="store",
dest="job_type",
help="job type: predict, extract\
predict: predicting,\
extract: extract features")
parser.add_option("-c", "--conf",
action="store", dest="train_conf",
help="network config")
parser.add_option("-i", "--data",
action="store", dest="data_file",
help="image list")
parser.add_option("-w", "--model",
action="store", dest="model_path",
default=None, help="model path")
parser.add_option("-g", "--use_gpu", action="store",
dest="use_gpu", default=True,
help="Whether to use gpu mode.")
parser.add_option("-o", "--output_dir",
action="store", dest="output_dir",
default="output", help="output path")
parser.add_option("-m", "--mean", action="store",
dest="mean", default=None,
help="mean file.")
parser.add_option("-p", "--multi_crop", action="store_true",
dest="multi_crop", default=False,
help="Wether to use multiple crops on image.")
parser.add_option(
"-c",
"--conf",
action="store",
dest="train_conf",
help="network config")
parser.add_option(
"-i", "--data", action="store", dest="data_file", help="image list")
parser.add_option(
"-w",
"--model",
action="store",
dest="model_path",
default=None,
help="model path")
parser.add_option(
"-g",
"--use_gpu",
action="store",
dest="use_gpu",
default=True,
help="Whether to use gpu mode.")
parser.add_option(
"-o",
"--output_dir",
action="store",
dest="output_dir",
default="output",
help="output path")
parser.add_option(
"-m",
"--mean",
action="store",
dest="mean",
default=None,
help="mean file.")
parser.add_option(
"-p",
"--multi_crop",
action="store_true",
dest="multi_crop",
default=False,
help="Wether to use multiple crops on image.")
parser.add_option("-l", "--output_layer", action="store",
dest="output_layer", default=None,
help="--job=extract, specify layers to extract "\
......@@ -254,24 +287,26 @@ def option_parser():
"classification probability, output in resnet.py.")
return parser.parse_args()
def main():
"""
1. parse input arguments.
2. predicting or extract features according job type.
"""
options, args = option_parser()
obj = ImageClassifier(options.train_conf,
options.model_path,
use_gpu=options.use_gpu,
mean_file=options.mean,
output_layer=options.output_layer,
oversample=options.multi_crop)
obj = ImageClassifier(
options.train_conf,
options.model_path,
use_gpu=options.use_gpu,
mean_file=options.mean,
output_layer=options.output_layer,
oversample=options.multi_crop)
if options.job_type == "predict":
obj.predict(options.data_file)
elif options.job_type == "extract":
obj.extract(options.data_file,
options.output_dir)
obj.extract(options.data_file, options.output_dir)
if __name__ == '__main__':
main()
demo/model_zoo/resnet/example/__init__.py
浏览文件 @ 84cab2c7
......@@ -11,4 +11,3 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
demo/model_zoo/resnet/example/image_list_provider.py
浏览文件 @ 84cab2c7
......@@ -16,8 +16,7 @@ from paddle.utils.image_util import *
from paddle.trainer.PyDataProvider2 import *
def hook(settings, image_size, crop_size, color, file_list,
is_train, **kwargs):
def hook(settings, image_size, crop_size, color, file_list, is_train, **kwargs):
"""
Description: Init with a list of data file
file_list is the name list of input files.
......@@ -58,7 +57,7 @@ def hook(settings, image_size, crop_size, color, file_list,
sz = settings.crop_size * settings.crop_size
settings.img_mean = np.zeros(sz * 3, dtype=np.single)
for idx, value in enumerate(settings.mean_value):
settings.img_mean[idx * sz: (idx + 1) * sz] = value
settings.img_mean[idx * sz:(idx + 1) * sz] = value
settings.img_mean = settings.img_mean.reshape(3, settings.crop_size,
settings.crop_size)
......@@ -69,7 +68,8 @@ def hook(settings, image_size, crop_size, color, file_list,
settings.input_types = [
dense_vector(settings.img_input_size), # image feature
integer_value(1)] # labels
integer_value(1)
] # labels
settings.logger.info('Image short side: %s', settings.img_size)
settings.logger.info('Crop size: %s', settings.crop_size)
......@@ -97,9 +97,6 @@ def processData(settings, file_list):
# swap channel
if settings.is_swap_channel:
img = img[settings.swap_channel, :, :]
img_feat = preprocess_img(img,
settings.img_mean,
settings.crop_size,
settings.is_train,
settings.color)
img_feat = preprocess_img(img, settings.img_mean, settings.crop_size,
settings.is_train, settings.color)
yield img_feat.tolist(), int(lab.strip())
demo/model_zoo/resnet/load_feature.py
浏览文件 @ 84cab2c7
......@@ -17,9 +17,11 @@ import sys
import cPickle
import logging
logging.basicConfig(format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s')
logging.basicConfig(
format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s')
logging.getLogger().setLevel(logging.INFO)
def load_feature_c(file):
"""
Load feature extracted by C++ interface.
......@@ -30,14 +32,15 @@ def load_feature_c(file):
f = open(file, 'r')
for line in f:
sample = []
for slot in line.strip().split(";"):
fea = [float(val) for val in slot.strip().split()]
for slot in line.strip().split(";"):
fea = [float(val) for val in slot.strip().split()]
if fea:
sample.append(fea)
features.append(sample)
f.close()
return features
def load_feature_py(feature_dir):
"""
Load feature extracted by python interface.
......@@ -54,6 +57,7 @@ def load_feature_py(feature_dir):
logging.info('Load feature file %s', file_name)
return features
if __name__ == '__main__':
print load_feature_py(sys.argv[1])
print load_feature_py(sys.argv[1])
#print load_feature_c(sys.argv[1])
demo/model_zoo/resnet/resnet.py
浏览文件 @ 84cab2c7
......@@ -13,7 +13,6 @@
# limitations under the License.
from paddle.trainer_config_helpers import *
"""
paper: https://arxiv.org/abs/1512.03385
"""
......@@ -28,15 +27,19 @@ if not is_predict and data_provider:
# mean.meta size : 3 x 224 x 224.
# If you use three mean value, set like:
# "mean_value:103.939,116.779,123.68;"
args={
args = {
'mean_meta': "model/mean_meta_224/mean.meta",
'image_size': 224, 'crop_size': 224,
'color': True,'swap_channel:': [2, 1, 0]}
define_py_data_sources2(train_list,
'example/test.list',
module="example.image_list_provider",
obj="processData",
args=args)
'image_size': 224,
'crop_size': 224,
'color': True,
'swap_channel:': [2, 1, 0]
}
define_py_data_sources2(
train_list,
'example/test.list',
module="example.image_list_provider",
obj="processData",
args=args)
batch_size = 1
learning_rate = 0.1 / batch_size
......@@ -54,12 +57,16 @@ Settings(
learning_method='momentum',
learning_rate_decay_a=0.5,
learning_rate_decay_b=1200000 * 10,
learning_rate_schedule="discexp",
)
learning_rate_schedule="discexp", )
def conv_bn_layer(name, input, filter_size, num_filters,
stride, padding, channels=None,
def conv_bn_layer(name,
input,
filter_size,
num_filters,
stride,
padding,
channels=None,
active_type=ReluActivation()):
"""
A wrapper for conv layer with batch normalization layers.
......@@ -67,19 +74,18 @@ def conv_bn_layer(name, input, filter_size, num_filters,
conv layer has no activation.
"""
tmp = img_conv_layer(name=name + "_conv",
input=input,
filter_size=filter_size,
num_channels=channels,
num_filters=num_filters,
stride=stride,
padding=padding,
act=LinearActivation(),
bias_attr=False)
return batch_norm_layer(name=name + "_bn",
input=tmp,
act=active_type,
use_global_stats=is_test)
tmp = img_conv_layer(
name=name + "_conv",
input=input,
filter_size=filter_size,
num_channels=channels,
num_filters=num_filters,
stride=stride,
padding=padding,
act=LinearActivation(),
bias_attr=False)
return batch_norm_layer(
name=name + "_bn", input=tmp, act=active_type, use_global_stats=is_test)
def bottleneck_block(name, input, num_filters1, num_filters2):
......@@ -88,29 +94,31 @@ def bottleneck_block(name, input, num_filters1, num_filters2):
Last conv_bn_layer has no activation.
Addto layer has activation of relu.
"""
last_name = conv_bn_layer(name=name + '_branch2a',
input=input,
filter_size=1,
num_filters=num_filters1,
stride=1,
padding=0)
last_name = conv_bn_layer(name=name + '_branch2b',
input=last_name,
filter_size=3,
num_filters=num_filters1,
stride=1,
padding=1)
last_name = conv_bn_layer(name=name + '_branch2c',
input=last_name,
filter_size=1,
num_filters=num_filters2,
stride=1,
padding=0,
active_type=LinearActivation())
return addto_layer(name=name + "_addto",
input=[input, last_name],
act=ReluActivation())
last_name = conv_bn_layer(
name=name + '_branch2a',
input=input,
filter_size=1,
num_filters=num_filters1,
stride=1,
padding=0)
last_name = conv_bn_layer(
name=name + '_branch2b',
input=last_name,
filter_size=3,
num_filters=num_filters1,
stride=1,
padding=1)
last_name = conv_bn_layer(
name=name + '_branch2c',
input=last_name,
filter_size=1,
num_filters=num_filters2,
stride=1,
padding=0,
active_type=LinearActivation())
return addto_layer(
name=name + "_addto", input=[input, last_name], act=ReluActivation())
def mid_projection(name, input, num_filters1, num_filters2, stride=2):
......@@ -123,38 +131,41 @@ def mid_projection(name, input, num_filters1, num_filters2, stride=2):
branch2x: bottleneck building block, shortcuts are identity.
"""
# stride = 2
branch1 = conv_bn_layer(name=name + '_branch1',
input=input,
filter_size=1,
num_filters=num_filters2,
stride=stride,
padding=0,
active_type=LinearActivation())
last_name = conv_bn_layer(name=name + '_branch2a',
input=input,
filter_size=1,
num_filters=num_filters1,
stride=stride,
padding=0)
last_name = conv_bn_layer(name=name + '_branch2b',
input=last_name,
filter_size=3,
num_filters=num_filters1,
stride=1,
padding=1)
last_name = conv_bn_layer(name=name + '_branch2c',
input=last_name,
filter_size=1,
num_filters=num_filters2,
stride=1,
padding=0,
active_type=LinearActivation())
return addto_layer(name=name + "_addto",
input=[branch1, last_name],
act=ReluActivation())
branch1 = conv_bn_layer(
name=name + '_branch1',
input=input,
filter_size=1,
num_filters=num_filters2,
stride=stride,
padding=0,
active_type=LinearActivation())
last_name = conv_bn_layer(
name=name + '_branch2a',
input=input,
filter_size=1,
num_filters=num_filters1,
stride=stride,
padding=0)
last_name = conv_bn_layer(
name=name + '_branch2b',
input=last_name,
filter_size=3,
num_filters=num_filters1,
stride=1,
padding=1)
last_name = conv_bn_layer(
name=name + '_branch2c',
input=last_name,
filter_size=1,
num_filters=num_filters2,
stride=1,
padding=0,
active_type=LinearActivation())
return addto_layer(
name=name + "_addto", input=[branch1, last_name], act=ReluActivation())
def deep_res_net(res2_num=3, res3_num=4, res4_num=6, res5_num=3):
......@@ -168,67 +179,67 @@ def deep_res_net(res2_num=3, res3_num=4, res4_num=6, res5_num=3):
# For ImageNet
# conv1: 112x112
img = data_layer(name='input', size=224 * 224 * 3)
tmp = conv_bn_layer("conv1", img,
filter_size=7,
channels=3,
num_filters=64,
stride=2,
padding=3)
tmp = conv_bn_layer(
"conv1",
img,
filter_size=7,
channels=3,
num_filters=64,
stride=2,
padding=3)
tmp = img_pool_layer(name="pool1", input=tmp, pool_size=3, stride=2)
# conv2_x: 56x56
tmp = mid_projection(name="res2_1",
input=tmp,
num_filters1=64,
num_filters2=256,
stride=1)
tmp = mid_projection(
name="res2_1", input=tmp, num_filters1=64, num_filters2=256, stride=1)
for i in xrange(2, res2_num + 1, 1):
tmp = bottleneck_block(name="res2_" + str(i),
input=tmp,
num_filters1=64,
num_filters2=256)
tmp = bottleneck_block(
name="res2_" + str(i), input=tmp, num_filters1=64, num_filters2=256)
# conv3_x: 28x28
tmp = mid_projection(name="res3_1",
input=tmp,
num_filters1=128,
num_filters2=512)
tmp = mid_projection(
name="res3_1", input=tmp, num_filters1=128, num_filters2=512)
for i in xrange(2, res3_num + 1, 1):
tmp = bottleneck_block(name="res3_" + str(i),
input=tmp, num_filters1=128,
num_filters2=512)
tmp = bottleneck_block(
name="res3_" + str(i),
input=tmp,
num_filters1=128,
num_filters2=512)
# conv4_x: 14x14
tmp = mid_projection(name="res4_1", input=tmp,
num_filters1=256, num_filters2=1024)
tmp = mid_projection(
name="res4_1", input=tmp, num_filters1=256, num_filters2=1024)
for i in xrange(2, res4_num + 1, 1):
tmp = bottleneck_block(name="res4_" + str(i),
input=tmp,
num_filters1=256,
num_filters2=1024)
tmp = bottleneck_block(
name="res4_" + str(i),
input=tmp,
num_filters1=256,
num_filters2=1024)
# conv5_x: 7x7
tmp = mid_projection(name="res5_1", input=tmp,
num_filters1=512, num_filters2=2048)
tmp = mid_projection(
name="res5_1", input=tmp, num_filters1=512, num_filters2=2048)
for i in xrange(2, res5_num + 1, 1):
tmp = bottleneck_block(name="res5_" + str(i),
input=tmp, num_filters1=512,
num_filters2=2048)
tmp = img_pool_layer(name='avgpool',
input=tmp,
pool_size=7,
stride=1,
pool_type=AvgPooling())
output = fc_layer(name='output',
input=tmp,
size=1000,
act=SoftmaxActivation())
tmp = bottleneck_block(
name="res5_" + str(i),
input=tmp,
num_filters1=512,
num_filters2=2048)
tmp = img_pool_layer(
name='avgpool',
input=tmp,
pool_size=7,
stride=1,
pool_type=AvgPooling())
output = fc_layer(
name='output', input=tmp, size=1000, act=SoftmaxActivation())
if not is_predict:
classification_cost(input=output, label=data_layer(name='label',
size=1))
classification_cost(
input=output, label=data_layer(
name='label', size=1))
def res_net_50():
......
demo/quick_start/api_train.py
浏览文件 @ 84cab2c7
......@@ -22,27 +22,32 @@ from py_paddle import DataProviderConverter
from paddle.trainer.PyDataProvider2 \
import integer_value, integer_value_sequence, sparse_binary_vector
def parse_arguments():
parser = argparse.ArgumentParser()
parser.add_argument("--train_data",
type=str, required=False, help="train data file")
parser.add_argument(
"--train_data", type=str, required=False, help="train data file")
parser.add_argument("--test_data", type=str, help="test data file")
parser.add_argument("--config",
type=str, required=True, help="config file name")
parser.add_argument(
"--config", type=str, required=True, help="config file name")
parser.add_argument("--dict_file", required=True, help="dictionary file")
parser.add_argument("--seq",
default=1, type=int,
help="whether use sequence training")
parser.add_argument("--use_gpu", default=0, type=int,
help="whether use GPU for training")
parser.add_argument("--trainer_count", default=1, type=int,
help="Number of threads for training")
parser.add_argument("--num_passes", default=5, type=int,
help="Number of training passes")
parser.add_argument(
"--seq", default=1, type=int, help="whether use sequence training")
parser.add_argument(
"--use_gpu", default=0, type=int, help="whether use GPU for training")
parser.add_argument(
"--trainer_count",
default=1,
type=int,
help="Number of threads for training")
parser.add_argument(
"--num_passes", default=5, type=int, help="Number of training passes")
return parser.parse_args()
UNK_IDX = 0
def load_data(file_name, word_dict):
with open(file_name, 'r') as f:
for line in f:
......@@ -51,6 +56,7 @@ def load_data(file_name, word_dict):
word_slot = [word_dict.get(w, UNK_IDX) for w in words]
yield word_slot, int(label)
def load_dict(dict_file):
word_dict = dict()
with open(dict_file, 'r') as f:
......@@ -59,6 +65,7 @@ def load_dict(dict_file):
word_dict[w] = i
return word_dict
def main():
options = parse_arguments()
api.initPaddle("--use_gpu=%s" % options.use_gpu,
......@@ -86,9 +93,9 @@ def main():
# create a data converter which converts data to PaddlePaddle
# internal format
input_types = [
integer_value_sequence(len(word_dict)) if options.seq
else sparse_binary_vector(len(word_dict)),
integer_value(2)]
integer_value_sequence(len(word_dict)) if options.seq else
sparse_binary_vector(len(word_dict)), integer_value(2)
]
converter = DataProviderConverter(input_types)
batch_size = trainer_config.opt_config.batch_size
......@@ -102,7 +109,7 @@ def main():
trainer.trainOneDataBatch(size, converter(batch))
trainer.finishTrainPass()
if test_dataset:
trainer.startTestPeriod();
trainer.startTestPeriod()
for pos in xrange(0, len(test_dataset), batch_size):
batch = itertools.islice(test_dataset, pos, pos + batch_size)
size = min(batch_size, len(test_dataset) - pos)
......@@ -110,5 +117,6 @@ def main():
trainer.finishTestPeriod()
trainer.finishTrain()
if __name__ == '__main__':
main()
demo/quick_start/dataprovider_bow.py
浏览文件 @ 84cab2c7
......@@ -17,6 +17,7 @@ from paddle.trainer.PyDataProvider2 import *
# id of the word not in dictionary
UNK_IDX = 0
# initializer is called by the framework during initialization.
# It allows the user to describe the data types and setup the
# necessary data structure for later use.
......@@ -38,7 +39,9 @@ def initializer(settings, dictionary, **kwargs):
# The second input is an integer. It represents the category id of the
# sample. 2 means there are two labels in the dataset.
# (1 for positive and 0 for negative)
integer_value(2)]
integer_value(2)
]
# Delaring a data provider. It has an initializer 'data_initialzer'.
# It will cache the generated data of the first pass in memory, so that
......@@ -69,9 +72,8 @@ def process(settings, file_name):
def predict_initializer(settings, dictionary, **kwargs):
settings.word_dict = dictionary
settings.input_types = [
sparse_binary_vector(len(dictionary))
]
settings.input_types = [sparse_binary_vector(len(dictionary))]
# Declaring a data provider for prediction. The difference with process
# is that label is not generated.
......
demo/quick_start/dataprovider_emb.py
浏览文件 @ 84cab2c7
......@@ -24,7 +24,8 @@ def initializer(settings, dictionary, **kwargs):
# The value of the integers range from 0 to len(dictrionary)-1
integer_value_sequence(len(dictionary)),
# Define the second input for label id
integer_value(2)]
integer_value(2)
]
@provider(init_hook=initializer, cache=CacheType.CACHE_PASS_IN_MEM)
......@@ -40,7 +41,8 @@ def process(settings, file_name):
def predict_initializer(settings, dictionary, **kwargs):
settings.word_dict = dictionary
settings.input_types = [
integer_value(len(dictionary), seq_type=SequenceType.SEQUENCE)
integer_value(
len(dictionary), seq_type=SequenceType.SEQUENCE)
]
......
demo/quick_start/preprocess.py
浏览文件 @ 84cab2c7
......@@ -13,7 +13,6 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
1. (remove HTML before or not)tokensizing
2. pos sample : rating score 5; neg sample: rating score 1-2.
......@@ -35,7 +34,8 @@ import multiprocessing
batch_size = 5000
word_count = {}
num_tokenize = max(1, multiprocessing.cpu_count() - 2) # parse + tokenize + save
num_tokenize = max(1,
multiprocessing.cpu_count() - 2) # parse + tokenize + save
max_queue_size = 8
parse_queue = Queue(maxsize=max_queue_size + num_tokenize)
tokenize_queue = Queue(maxsize=max_queue_size + num_tokenize)
......
demo/quick_start/train.sh
浏览文件 @ 84cab2c7
......@@ -20,6 +20,7 @@ cfg=trainer_config.lr.py
#cfg=trainer_config.lstm.py
#cfg=trainer_config.bidi-lstm.py
#cfg=trainer_config.db-lstm.py
#cfg=trainer_config.resnet-lstm.py
paddle train \
--config=$cfg \
--save_dir=./output \
......
demo/quick_start/trainer_config.bidi-lstm.py
浏览文件 @ 84cab2c7
......@@ -27,11 +27,12 @@ is_predict = get_config_arg('is_predict', bool, False)
trn = 'data/train.list' if not is_predict else None
tst = 'data/test.list' if not is_predict else 'data/pred.list'
process = 'process' if not is_predict else 'process_predict'
define_py_data_sources2(train_list=trn,
test_list=tst,
module="dataprovider_emb",
obj=process,
args={"dictionary": word_dict})
define_py_data_sources2(
train_list=trn,
test_list=tst,
module="dataprovider_emb",
obj=process,
args={"dictionary": word_dict})
batch_size = 128 if not is_predict else 1
settings(
......@@ -39,19 +40,17 @@ settings(
learning_rate=2e-3,
learning_method=AdamOptimizer(),
regularization=L2Regularization(8e-4),
gradient_clipping_threshold=25
)
gradient_clipping_threshold=25)
bias_attr = ParamAttr(initial_std=0.,l2_rate=0.)
bias_attr = ParamAttr(initial_std=0., l2_rate=0.)
data = data_layer(name="word", size=len(word_dict))
emb = embedding_layer(input=data, size=128)
bi_lstm = bidirectional_lstm(input=emb, size=128)
dropout = dropout_layer(input=bi_lstm, dropout_rate=0.5)
output = fc_layer(input=dropout, size=2,
bias_attr=bias_attr,
act=SoftmaxActivation())
output = fc_layer(
input=dropout, size=2, bias_attr=bias_attr, act=SoftmaxActivation())
if is_predict:
maxid = maxid_layer(output)
......
demo/quick_start/trainer_config.cnn.py
浏览文件 @ 84cab2c7
......@@ -27,11 +27,12 @@ is_predict = get_config_arg('is_predict', bool, False)
trn = 'data/train.list' if not is_predict else None
tst = 'data/test.list' if not is_predict else 'data/pred.list'
process = 'process' if not is_predict else 'process_predict'
define_py_data_sources2(train_list=trn,
test_list=tst,
module="dataprovider_emb",
obj=process,
args={"dictionary": word_dict})
define_py_data_sources2(
train_list=trn,
test_list=tst,
module="dataprovider_emb",
obj=process,
args={"dictionary": word_dict})
batch_size = 128 if not is_predict else 1
settings(
......@@ -39,8 +40,7 @@ settings(
learning_rate=2e-3,
learning_method=AdamOptimizer(),
regularization=L2Regularization(8e-4),
gradient_clipping_threshold=25
)
gradient_clipping_threshold=25)
data = data_layer(name="word", size=len(word_dict))
embedding = embedding_layer(input=data, size=128)
......
demo/quick_start/trainer_config.db-lstm.py
浏览文件 @ 84cab2c7
......@@ -27,11 +27,12 @@ is_predict = get_config_arg('is_predict', bool, False)
trn = 'data/train.list' if not is_predict else None
tst = 'data/test.list' if not is_predict else 'data/pred.list'
process = 'process' if not is_predict else 'process_predict'
define_py_data_sources2(train_list=trn,
test_list=tst,
module="dataprovider_emb",
obj=process,
args={"dictionary": word_dict})
define_py_data_sources2(
train_list=trn,
test_list=tst,
module="dataprovider_emb",
obj=process,
args={"dictionary": word_dict})
batch_size = 128 if not is_predict else 1
settings(
......@@ -39,10 +40,9 @@ settings(
learning_rate=2e-3,
learning_method=AdamOptimizer(),
regularization=L2Regularization(8e-4),
gradient_clipping_threshold=25
)
gradient_clipping_threshold=25)
bias_attr = ParamAttr(initial_std=0.,l2_rate=0.)
bias_attr = ParamAttr(initial_std=0., l2_rate=0.)
data = data_layer(name="word", size=len(word_dict))
emb = embedding_layer(input=data, size=128)
......@@ -52,17 +52,18 @@ lstm_0 = lstmemory(input=hidden_0, layer_attr=ExtraAttr(drop_rate=0.1))
input_layers = [hidden_0, lstm_0]
for i in range(1,8):
for i in range(1, 8):
fc = fc_layer(input=input_layers, size=128)
lstm = lstmemory(input=fc, layer_attr=ExtraAttr(drop_rate=0.1),
reverse=(i % 2) == 1,)
lstm = lstmemory(
input=fc,
layer_attr=ExtraAttr(drop_rate=0.1),
reverse=(i % 2) == 1, )
input_layers = [fc, lstm]
lstm_last = pooling_layer(input=lstm, pooling_type=MaxPooling())
output = fc_layer(input=lstm_last, size=2,
bias_attr=bias_attr,
act=SoftmaxActivation())
output = fc_layer(
input=lstm_last, size=2, bias_attr=bias_attr, act=SoftmaxActivation())
if is_predict:
maxid = maxid_layer(output)
......
demo/quick_start/trainer_config.emb.py
浏览文件 @ 84cab2c7
......@@ -27,18 +27,16 @@ is_predict = get_config_arg('is_predict', bool, False)
trn = 'data/train.list' if not is_predict else None
tst = 'data/test.list' if not is_predict else 'data/pred.list'
process = 'process' if not is_predict else 'process_predict'
define_py_data_sources2(train_list=trn,
test_list=tst,
module="dataprovider_emb",
obj=process,
args={"dictionary": word_dict})
define_py_data_sources2(
train_list=trn,
test_list=tst,
module="dataprovider_emb",
obj=process,
args={"dictionary": word_dict})
batch_size = 128 if not is_predict else 1
settings(
batch_size=batch_size,
learning_rate=2e-3,
learning_method=AdamOptimizer()
)
batch_size=batch_size, learning_rate=2e-3, learning_method=AdamOptimizer())
data = data_layer(name="word", size=len(word_dict))
embedding = embedding_layer(input=data, size=128)
......
demo/quick_start/trainer_config.lr.py
浏览文件 @ 84cab2c7
......@@ -32,11 +32,12 @@ process = 'process' if not is_predict else 'process_predict'
# We need to use different process for training and prediction.
# For training, the input data includes both word IDs and labels.
# For prediction, the input data only includs word Ids.
define_py_data_sources2(train_list=trn,
test_list=tst,
module="dataprovider_bow",
obj=process,
args={"dictionary": word_dict})
define_py_data_sources2(
train_list=trn,
test_list=tst,
module="dataprovider_bow",
obj=process,
args={"dictionary": word_dict})
batch_size = 128 if not is_predict else 1
settings(
......@@ -44,8 +45,7 @@ settings(
learning_rate=2e-3,
learning_method=AdamOptimizer(),
regularization=L2Regularization(8e-4),
gradient_clipping_threshold=25
)
gradient_clipping_threshold=25)
# Define the data for text features. The size of the data layer is the number
# of words in the dictionary.
......
demo/quick_start/trainer_config.lstm.py
浏览文件 @ 84cab2c7
......@@ -27,11 +27,12 @@ is_predict = get_config_arg('is_predict', bool, False)
trn = 'data/train.list' if not is_predict else None
tst = 'data/test.list' if not is_predict else 'data/pred.list'
process = 'process' if not is_predict else 'process_predict'
define_py_data_sources2(train_list=trn,
test_list=tst,
module="dataprovider_emb",
obj=process,
args={"dictionary": word_dict})
define_py_data_sources2(
train_list=trn,
test_list=tst,
module="dataprovider_emb",
obj=process,
args={"dictionary": word_dict})
batch_size = 128 if not is_predict else 1
settings(
......@@ -39,17 +40,14 @@ settings(
learning_rate=2e-3,
learning_method=AdamOptimizer(),
regularization=L2Regularization(8e-4),
gradient_clipping_threshold=25
)
gradient_clipping_threshold=25)
data = data_layer(name="word", size=len(word_dict))
emb = embedding_layer(input=data, size=128)
lstm = simple_lstm(input=emb, size=128,
lstm_cell_attr=ExtraAttr(drop_rate=0.25))
lstm = simple_lstm(
input=emb, size=128, lstm_cell_attr=ExtraAttr(drop_rate=0.25))
lstm_max = pooling_layer(input=lstm, pooling_type=MaxPooling())
output = fc_layer(input=lstm_max, size=2,
act=SoftmaxActivation())
output = fc_layer(input=lstm_max, size=2, act=SoftmaxActivation())
if is_predict:
maxid = maxid_layer(output)
outputs([maxid, output])
......
demo/quick_start/trainer_config.resnet-lstm.py 0 → 100644
浏览文件 @ 84cab2c7
# edit-mode: -*- python -*-
# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
This configuration is a demonstration of how to implement the stacked LSTM
with residual connections, i.e. an LSTM layer takes the sum of the hidden states
and inputs of the previous LSTM layer instead of only the hidden states.
This architecture is from:
Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V. Le, Mohammad Norouzi,
Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey,
Jeff Klingner, Apurva Shah, Melvin Johnson, Xiaobing Liu, Lukasz Kaiser,
Stephan Gouws, Yoshikiyo Kato, Taku Kudo, Hideto Kazawa, Keith Stevens,
George Kurian, Nishant Patil, Wei Wang, Cliff Young, Jason Smith, Jason Riesa,
Alex Rudnick, Oriol Vinyals, Greg Corrado, Macduff Hughes, Jeffrey Dean. 2016.
Google's Neural Machine Translation System: Bridging the Gap between Human and
Machine Translation. In arXiv https://arxiv.org/pdf/1609.08144v2.pdf
Different from the architecture described in the paper, we use a stack single
direction LSTM layers as the first layer instead of bi-directional LSTM. Also,
since this is a demo code, to reduce computation time, we stacked 4 layers
instead of 8 layers.
"""
from paddle.trainer_config_helpers import *
dict_file = "./data/dict.txt"
word_dict = dict()
with open(dict_file, 'r') as f:
for i, line in enumerate(f):
w = line.strip().split()[0]
word_dict[w] = i
is_predict = get_config_arg('is_predict', bool, False)
trn = 'data/train.list' if not is_predict else None
tst = 'data/test.list' if not is_predict else 'data/pred.list'
process = 'process' if not is_predict else 'process_predict'
define_py_data_sources2(train_list=trn,
test_list=tst,
module="dataprovider_emb",
obj=process,
args={"dictionary": word_dict})
batch_size = 128 if not is_predict else 1
settings(
batch_size=batch_size,
learning_rate=2e-3,
learning_method=AdamOptimizer(),
regularization=L2Regularization(8e-4),
gradient_clipping_threshold=25
)
bias_attr = ParamAttr(initial_std=0.,l2_rate=0.)
data = data_layer(name="word", size=len(word_dict))
emb = embedding_layer(input=data, size=128)
lstm = simple_lstm(input=emb, size=128, lstm_cell_attr=ExtraAttr(drop_rate=0.1))
previous_input, previous_hidden_state = emb, lstm
for i in range(3):
# The input to the current layer is the sum of the hidden state
# and input of the previous layer.
current_input = addto_layer(input=[previous_input, previous_hidden_state])
hidden_state = simple_lstm(input=current_input, size=128,
lstm_cell_attr=ExtraAttr(drop_rate=0.1))
previous_input, previous_hidden_state = current_input, hidden_state
lstm = previous_hidden_state
lstm_last = pooling_layer(input=lstm, pooling_type=MaxPooling())
output = fc_layer(input=lstm_last, size=2,
bias_attr=bias_attr,
act=SoftmaxActivation())
if is_predict:
maxid = maxid_layer(output)
outputs([maxid, output])
else:
label = data_layer(name="label", size=2)
cls = classification_cost(input=output, label=label)
outputs(cls)
demo/recommendation/common_utils.py
浏览文件 @ 84cab2c7
......@@ -21,8 +21,9 @@ def meta_to_header(meta, name):
yield integer_value(each_meta['max'])
elif each_meta['type'] == 'embedding':
is_seq = each_meta['seq'] == 'sequence'
yield integer_value(len(each_meta['dict']),
seq_type=SequenceType.SEQUENCE if is_seq
else SequenceType.NO_SEQUENCE)
yield integer_value(
len(each_meta['dict']),
seq_type=SequenceType.SEQUENCE
if is_seq else SequenceType.NO_SEQUENCE)
elif each_meta['type'] == 'one_hot_dense':
yield dense_vector(len(each_meta['dict']))
demo/recommendation/data/config.json
浏览文件 @ 84cab2c7
......@@ -14,4 +14,3 @@
"fields": ["id", "title", "genres"]
}
}
demo/recommendation/data/config_generator.py
浏览文件 @ 84cab2c7
......@@ -12,7 +12,6 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
config_generator.py
......@@ -29,10 +28,7 @@ import json
import docopt
import copy
DEFAULT_FILE = {
"type": "split",
"delimiter": ","
}
DEFAULT_FILE = {"type": "split", "delimiter": ","}
DEFAULT_FIELD = {
"id": {
......@@ -107,19 +103,16 @@ def main(filename, fmt):
field = copy.deepcopy(DEFAULT_FIELD[field_key])
field['pos'] = pos
fields.append(field)
obj[k] = {
"file": file_dict,
"fields": fields
}
meta = {
"meta": obj
}
obj[k] = {"file": file_dict, "fields": fields}
meta = {"meta": obj}
# print meta
if fmt == 'json':
def formatter(x):
import json
return json.dumps(x, indent=2)
elif fmt == 'yaml':
def formatter(x):
import yaml
return yaml.safe_dump(x, default_flow_style=False)
......
demo/recommendation/data/meta_generator.py
浏览文件 @ 84cab2c7
......@@ -12,7 +12,6 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Preprocess Movielens dataset, to get movie/user object.
......@@ -66,8 +65,8 @@ class SortedIDGenerator(object):
self.__key_set__.add(key)
def finish_scan(self, compare=None, key=None, reverse=False):
self.__key_set__ = sorted(list(self.__key_set__), cmp=compare,
key=key, reverse=reverse)
self.__key_set__ = sorted(
list(self.__key_set__), cmp=compare, key=key, reverse=reverse)
self.dict = dict()
for idx, each_key in enumerate(self.__key_set__):
self.dict[each_key] = idx
......@@ -207,11 +206,10 @@ class EmbeddingFieldParser(object):
self.dict = EmbeddingFieldParser.CharBasedEmbeddingDict(
self.seq_type == EmbeddingFieldParser.SEQUENCE)
elif config['dict']['type'] == 'split':
self.dict = SplitEmbeddingDict(
config['dict'].get('delimiter', ','))
self.dict = SplitEmbeddingDict(config['dict'].get('delimiter', ','))
elif config['dict']['type'] == 'whole_content':
self.dict = EmbeddingFieldParser.WholeContentDict(
config['dict']['sort'])
self.dict = EmbeddingFieldParser.WholeContentDict(config['dict'][
'sort'])
else:
print config
assert False
......@@ -333,8 +331,8 @@ class ContentExtractorFactory(object):
return PositionContentExtractor(config['pos'])
else:
extra_args = config['regex']
return RegexPositionContentExtractor(pos=config['pos'],
**extra_args)
return RegexPositionContentExtractor(
pos=config['pos'], **extra_args)
class MetaFile(object):
......@@ -364,9 +362,10 @@ class MetaFile(object):
metas = map(lambda x: x.meta_field(), field_parsers)
# print metas
key_index = filter(lambda x: x is not None, map(
lambda (idx, meta): idx if 'is_key' in meta and meta['is_key']
else None, enumerate(metas)))[0]
key_index = filter(
lambda x: x is not None,
map(lambda (idx, meta): idx if 'is_key' in meta and meta['is_key'] else None,
enumerate(metas)))[0]
key_map = []
for i in range(min(key_index, len(metas))):
......@@ -374,12 +373,7 @@ class MetaFile(object):
for i in range(key_index + 1, len(metas)):
key_map.append(i)
obj = {
'__meta__': {
'raw_meta': metas,
'feature_map': key_map
}
}
obj = {'__meta__': {'raw_meta': metas, 'feature_map': key_map}}
for each_block in reader.read():
idx = field_parsers[key_index].parse(each_block)
......
demo/recommendation/data/split.py
浏览文件 @ 84cab2c7
......@@ -12,7 +12,6 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Separate movielens 1m dataset to train/test file.
......
demo/recommendation/dataprovider.py
浏览文件 @ 84cab2c7
......@@ -15,6 +15,7 @@
from paddle.trainer.PyDataProvider2 import *
import common_utils # parse
def hook(settings, meta, **kwargs):
"""
Init hook is invoked before process data. It will set obj.slots and store
......@@ -41,6 +42,7 @@ def hook(settings, meta, **kwargs):
settings.input_types = headers
settings.meta = meta
@provider(init_hook=hook, cache=CacheType.CACHE_PASS_IN_MEM)
def process(settings, filename):
with open(filename, 'r') as f:
......
demo/recommendation/prediction.py
浏览文件 @ 84cab2c7
......@@ -28,7 +28,8 @@ if __name__ == '__main__':
model_path = sys.argv[1]
swig_paddle.initPaddle('--use_gpu=0')
conf = parse_config("trainer_config.py", "is_predict=1")
network = swig_paddle.GradientMachine.createFromConfigProto(conf.model_config)
network = swig_paddle.GradientMachine.createFromConfigProto(
conf.model_config)
assert isinstance(network, swig_paddle.GradientMachine)
network.loadParameters(model_path)
with open('./data/meta.bin', 'rb') as f:
......@@ -39,11 +40,12 @@ if __name__ == '__main__':
while True:
movie_id = int(raw_input("Input movie_id: "))
user_id = int(raw_input("Input user_id: "))
movie_meta = meta['movie'][movie_id] # Query Data From Meta.
movie_meta = meta['movie'][movie_id] # Query Data From Meta.
user_meta = meta['user'][user_id]
data = [movie_id - 1]
data.extend(movie_meta)
data.append(user_id - 1)
data.extend(user_meta)
print "Prediction Score is %.2f" % ((network.forwardTest(
cvt.convert([data]))[0]['value'][0][0] + 5) / 2)
print "Prediction Score is %.2f" % (
(network.forwardTest(cvt.convert([data]))[0]['value'][0][0] + 5)
/ 2)
demo/recommendation/trainer_config.py
浏览文件 @ 84cab2c7
......@@ -27,8 +27,8 @@ with open(META_FILE, 'rb') as f:
# load meta file
meta = pickle.load(f)
settings(batch_size=1600, learning_rate=1e-3,
learning_method=RMSPropOptimizer())
settings(
batch_size=1600, learning_rate=1e-3, learning_method=RMSPropOptimizer())
def construct_feature(name):
......@@ -59,11 +59,10 @@ def construct_feature(name):
slot_name = each_meta.get('name', '%s_id' % name)
if type_name == 'id':
slot_dim = each_meta['max']
embedding = embedding_layer(input=data_layer(slot_name,
size=slot_dim),
size=256)
fusion.append(fc_layer(input=embedding,
size=256))
embedding = embedding_layer(
input=data_layer(
slot_name, size=slot_dim), size=256)
fusion.append(fc_layer(input=embedding, size=256))
elif type_name == 'embedding':
is_seq = each_meta['seq'] == 'sequence'
slot_dim = len(each_meta['dict'])
......@@ -71,17 +70,14 @@ def construct_feature(name):
embedding = embedding_layer(input=din, size=256)
if is_seq:
fusion.append(
text_conv_pool(input=embedding, context_len=5,
hidden_size=256))
text_conv_pool(
input=embedding, context_len=5, hidden_size=256))
else:
fusion.append(fc_layer(input=embedding,
size=256))
fusion.append(fc_layer(input=embedding, size=256))
elif type_name == 'one_hot_dense':
slot_dim = len(each_meta['dict'])
hidden = fc_layer(input=data_layer(slot_name, slot_dim),
size=256)
fusion.append(fc_layer(input=hidden,
size=256))
hidden = fc_layer(input=data_layer(slot_name, slot_dim), size=256)
fusion.append(fc_layer(input=hidden, size=256))
return fc_layer(name="%s_fusion" % name, input=fusion, size=256)
......@@ -90,10 +86,16 @@ movie_feature = construct_feature("movie")
user_feature = construct_feature("user")
similarity = cos_sim(a=movie_feature, b=user_feature)
if not is_predict:
outputs(regression_cost(input=similarity,
label=data_layer('rating', size=1)))
define_py_data_sources2('data/train.list', 'data/test.list', module='dataprovider',
obj='process', args={'meta': meta})
outputs(
regression_cost(
input=similarity, label=data_layer(
'rating', size=1)))
define_py_data_sources2(
'data/train.list',
'data/test.list',
module='dataprovider',
obj='process',
args={'meta': meta})
else:
outputs(similarity)
demo/semantic_role_labeling/dataprovider.py
浏览文件 @ 84cab2c7
......@@ -26,9 +26,9 @@ def hook(settings, word_dict, label_dict, **kwargs):
integer_value_sequence(len(word_dict)),
integer_value_sequence(len(word_dict)),
integer_value_sequence(len(word_dict)),
integer_value_sequence(len(word_dict)),
integer_value_sequence(2),
integer_value_sequence(len(label_dict))]
integer_value_sequence(len(word_dict)), integer_value_sequence(2),
integer_value_sequence(len(label_dict))
]
@provider(init_hook=hook)
......
demo/semantic_role_labeling/db_lstm.py
浏览文件 @ 84cab2c7
......@@ -12,7 +12,6 @@
# See the License for the specific language governing permissions and
# limitations under the License.
import math
import os
import sys
......@@ -42,7 +41,7 @@ if not is_predict:
label_dict[w] = i
if is_test:
train_list_file = None
train_list_file = None
#define data provider
define_py_data_sources2(
......
demo/semantic_role_labeling/predict.py
浏览文件 @ 84cab2c7
......@@ -41,22 +41,16 @@ class Prediction():
len_dict = len(self.dict)
len_label = len(self.labels)
conf = parse_config(
train_conf,
'dict_len=' + str(len_dict) +
',label_len=' + str(len_label) +
',is_predict=True')
conf = parse_config(train_conf, 'dict_len=' + str(len_dict) +
',label_len=' + str(len_label) + ',is_predict=True')
self.network = swig_paddle.GradientMachine.createFromConfigProto(
conf.model_config)
self.network.loadParameters(model_dir)
slots = [
integer_value_sequence(len_dict),
integer_value_sequence(len_dict),
integer_value_sequence(len_dict),
integer_value_sequence(len_dict),
integer_value_sequence(len_dict),
integer_value_sequence(2)
integer_value_sequence(len_dict), integer_value_sequence(len_dict),
integer_value_sequence(len_dict), integer_value_sequence(len_dict),
integer_value_sequence(len_dict), integer_value_sequence(2)
]
self.converter = DataProviderConverter(slots)
......@@ -110,8 +104,8 @@ class Prediction():
len_sen = len(sen.split())
line_labels = lab[index:index + len_sen]
index += len_sen
fout.write(sen + '\t' + ' '.join([self.labels_reverse[
i] for i in line_labels]) + '\n')
fout.write(sen + '\t' + ' '.join(
[self.labels_reverse[i] for i in line_labels]) + '\n')
def option_parser():
......
demo/semantic_role_labeling/test.sh
浏览文件 @ 84cab2c7
......@@ -37,4 +37,3 @@ paddle train \
--use_gpu=false \
--config_args=is_test=1 \
2>&1 | tee 'test.log'
demo/semantic_role_labeling/train.sh
浏览文件 @ 84cab2c7
......@@ -24,4 +24,3 @@ paddle train \
--show_parameter_stats_period=10 \
--test_all_data_in_one_period=1 \
2>&1 | tee 'train.log'
demo/sentiment/data/get_imdb.sh
浏览文件 @ 84cab2c7
......@@ -38,11 +38,11 @@ unzip master.zip
mkdir -p imdb/train
mkdir -p imdb/test
cp -r aclImdb/train/pos/ imdb/train/
cp -r aclImdb/train/neg/ imdb/train/
cp -r aclImdb/train/pos/ imdb/train/pos
cp -r aclImdb/train/neg/ imdb/train/neg
cp -r aclImdb/test/pos/ imdb/test/
cp -r aclImdb/test/neg/ imdb/test/
cp -r aclImdb/test/pos/ imdb/test/pos
cp -r aclImdb/test/neg/ imdb/test/neg
#remove compressed package
rm aclImdb_v1.tar.gz
......
demo/sentiment/dataprovider.py
浏览文件 @ 84cab2c7
......@@ -17,8 +17,8 @@ from paddle.trainer.PyDataProvider2 import *
def hook(settings, dictionary, **kwargs):
settings.word_dict = dictionary
settings.input_types = [
integer_value_sequence(len(settings.word_dict)),
integer_value(2)]
integer_value_sequence(len(settings.word_dict)), integer_value(2)
]
settings.logger.info('dict len : %d' % (len(settings.word_dict)))
......@@ -29,6 +29,7 @@ def process(settings, file_name):
label, comment = line.strip().split('\t\t')
label = int(label)
words = comment.split()
word_slot = [settings.word_dict[w] for w in words if w in
settings.word_dict]
word_slot = [
settings.word_dict[w] for w in words if w in settings.word_dict
]
yield word_slot, label
demo/sentiment/predict.py
浏览文件 @ 84cab2c7
......@@ -18,14 +18,14 @@ from optparse import OptionParser
from py_paddle import swig_paddle, DataProviderConverter
from paddle.trainer.PyDataProvider2 import integer_value_sequence
from paddle.trainer.config_parser import parse_config
"""
Usage: run following command to show help message.
python predict.py -h
"""
class SentimentPrediction():
def __init__(self, train_conf, dict_file, model_dir=None, label_file = None):
def __init__(self, train_conf, dict_file, model_dir=None, label_file=None):
"""
train_conf: trainer configure.
dict_file: word dictionary file name.
......@@ -44,7 +44,8 @@ class SentimentPrediction():
self.load_label(label_file)
conf = parse_config(train_conf, "is_predict=1")
self.network = swig_paddle.GradientMachine.createFromConfigProto(conf.model_config)
self.network = swig_paddle.GradientMachine.createFromConfigProto(
conf.model_config)
self.network.loadParameters(self.model_dir)
input_types = [integer_value_sequence(self.dict_dim)]
self.converter = DataProviderConverter(input_types)
......@@ -61,7 +62,7 @@ class SentimentPrediction():
"""
Load label.
"""
self.label={}
self.label = {}
for v in open(label_file, 'r'):
self.label[int(v.split('\t')[1])] = v.split('\t')[0]
......@@ -72,7 +73,9 @@ class SentimentPrediction():
with open(data_file, 'r') as fdata:
for line in fdata:
words = line.strip().split()
word_slot = [self.word_dict[w] for w in words if w in self.word_dict]
word_slot = [
self.word_dict[w] for w in words if w in self.word_dict
]
if not word_slot:
print "all words are not in dictionary: %s", line
continue
......@@ -89,25 +92,48 @@ class SentimentPrediction():
if self.label is None:
print("%s: predicting label is %d" % (data_file, lab[0][0]))
else:
print("%s: predicting label is %s" % (data_file, self.label[lab[0][0]]))
print("%s: predicting label is %s" %
(data_file, self.label[lab[0][0]]))
def option_parser():
usage = "python predict.py -n config -w model_dir -d dictionary -i input_file "
parser = OptionParser(usage="usage: %s [options]" % usage)
parser.add_option("-n", "--tconf", action="store",
dest="train_conf", help="network config")
parser.add_option("-d", "--dict", action="store",
dest="dict_file",help="dictionary file")
parser.add_option("-b", "--label", action="store",
dest="label", default=None,
help="dictionary file")
parser.add_option("-i", "--data", action="store",
dest="data", help="data file to predict")
parser.add_option("-w", "--model", action="store",
dest="model_path", default=None,
help="model path")
parser.add_option(
"-n",
"--tconf",
action="store",
dest="train_conf",
help="network config")
parser.add_option(
"-d",
"--dict",
action="store",
dest="dict_file",
help="dictionary file")
parser.add_option(
"-b",
"--label",
action="store",
dest="label",
default=None,
help="dictionary file")
parser.add_option(
"-i",
"--data",
action="store",
dest="data",
help="data file to predict")
parser.add_option(
"-w",
"--model",
action="store",
dest="model_path",
default=None,
help="model path")
return parser.parse_args()
def main():
options, args = option_parser()
train_conf = options.train_conf
......@@ -119,5 +145,6 @@ def main():
predict = SentimentPrediction(train_conf, dict_file, model_path, label)
predict.predict(data)
if __name__ == '__main__':
main()
demo/sentiment/preprocess.py
浏览文件 @ 84cab2c7
......@@ -22,13 +22,13 @@ from os.path import join as join_path
from optparse import OptionParser
from paddle.utils.preprocess_util import *
"""
Usage: run following command to show help message.
python preprocess.py -h
"""
def save_dict(dict, filename, is_reverse = True):
def save_dict(dict, filename, is_reverse=True):
"""
Save dictionary into file.
dict: input dictionary.
......@@ -39,9 +39,10 @@ def save_dict(dict, filename, is_reverse = True):
f = open(filename, 'w')
for k, v in sorted(dict.items(), key=operator.itemgetter(1),\
reverse=is_reverse):
f.write('%s\t%s\n'%(k, v))
f.write('%s\t%s\n' % (k, v))
f.close()
def tokenize(sentences):
"""
Use tokenizer.perl to tokenize input sentences.
......@@ -58,6 +59,7 @@ def tokenize(sentences):
toks = tok_text.split('\n')[:-1]
return toks
def read_lines(path):
"""
path: String, file path.
......@@ -71,12 +73,17 @@ def read_lines(path):
seqs.append(line)
return seqs
class SentimentDataSetCreate():
"""
A class to process data for sentiment analysis task.
"""
def __init__(self, data_path, output_path,
use_okenizer = True, multi_lines = False):
def __init__(self,
data_path,
output_path,
use_okenizer=True,
multi_lines=False):
"""
data_path: string, traing and testing dataset path
output_path: string, output path, store processed dataset
......@@ -164,23 +171,17 @@ class SentimentDataSetCreate():
# Preprocess train data.
train_data, train_lab_set = self.data_list(self.train_dir)
print "processing train set..."
file_lists = self.save_data(train_data,
"train",
self.batch_size,
True,
True)
file_lists = self.save_data(train_data, "train", self.batch_size, True,
True)
save_list(file_lists, self.train_list)
# If have test data path, preprocess test data.
if os.path.exists(self.test_dir):
test_data, test_lab_set = self.data_list(self.test_dir)
assert(train_lab_set == test_lab_set)
assert (train_lab_set == test_lab_set)
print "processing test set..."
file_lists = self.save_data(test_data,
"test",
self.batch_size,
False,
self.dict_with_test)
file_lists = self.save_data(test_data, "test", self.batch_size,
False, self.dict_with_test)
save_list(file_lists, self.test_list)
# save labels set.
......@@ -191,7 +192,9 @@ class SentimentDataSetCreate():
save_dict(self.word_count, self.dict_file, True)
self.dict_size = len(self.word_count)
def save_data(self, data, prefix = "",
def save_data(self,
data,
prefix="",
batch_size=50000,
is_shuffle=False,
build_dict=False):
......@@ -205,7 +208,8 @@ class SentimentDataSetCreate():
return: list of batch names
"""
if is_shuffle and self.multi_lines:
return self.save_data_multi_lines(data, prefix, batch_size, build_dict)
return self.save_data_multi_lines(data, prefix, batch_size,
build_dict)
if is_shuffle:
random.shuffle(data)
......@@ -213,7 +217,7 @@ class SentimentDataSetCreate():
batch_names = []
for i in range(num_batches):
batch_name = join_path(self.output_path,
"%s_part_%03d" %(prefix, i))
"%s_part_%03d" % (prefix, i))
begin = i * batch_size
end = min((i + 1) * batch_size, len(data))
# read a batch of data
......@@ -246,7 +250,9 @@ class SentimentDataSetCreate():
data_list = tokenize(data_list)
return label_list, data_list
def save_data_multi_lines(self, data, prefix = "",
def save_data_multi_lines(self,
data,
prefix="",
batch_size=50000,
build_dict=False):
"""
......@@ -274,14 +280,14 @@ class SentimentDataSetCreate():
self.create_dict(data_list)
length = len(label_list)
perm_list = np.array([ i for i in xrange(length) ])
perm_list = np.array([i for i in xrange(length)])
random.shuffle(perm_list)
num_batches = int(math.ceil(length / float(batch_size)))
batch_names = []
for i in range(num_batches):
batch_name = join_path(self.output_path,
"%s_part_%03d" %(prefix, i))
"%s_part_%03d" % (prefix, i))
begin = i * batch_size
end = min((i + 1) * batch_size, length)
sub_label = [label_list[perm_list[i]] for i in range(begin, end)]
......@@ -304,35 +310,50 @@ class SentimentDataSetCreate():
f.write('%s\t\t%s\n' % (lab, seq))
f.close()
def option_parser():
parser = OptionParser(usage="usage: python preprcoess.py "\
"-i data_dir [options]")
parser.add_option("-i", "--data", action="store",
dest="input", help="Input data directory.")
parser.add_option("-o", "--output", action="store",
dest="output", default=None,
help="Output directory.")
parser.add_option("-t", "--tokenizer", action="store",
dest="use_tokenizer", default=True,
help="Whether to use tokenizer.")
parser.add_option(
"-i",
"--data",
action="store",
dest="input",
help="Input data directory.")
parser.add_option(
"-o",
"--output",
action="store",
dest="output",
default=None,
help="Output directory.")
parser.add_option(
"-t",
"--tokenizer",
action="store",
dest="use_tokenizer",
default=True,
help="Whether to use tokenizer.")
parser.add_option("-m", "--multi_lines", action="store",
dest="multi_lines", default=False,
help="If input text files have multi lines and they "\
"need to be shuffled, you should set -m True,")
return parser.parse_args()
def main():
options, args = option_parser()
data_dir=options.input
output_dir=options.output
use_tokenizer=options.use_tokenizer
multi_lines=options.multi_lines
data_dir = options.input
output_dir = options.output
use_tokenizer = options.use_tokenizer
multi_lines = options.multi_lines
if output_dir is None:
outname = os.path.basename(options.input)
output_dir = join_path(os.path.dirname(data_dir), 'pre-' + outname)
data_creator = SentimentDataSetCreate(data_dir, output_dir,
use_tokenizer, multi_lines)
data_creator = SentimentDataSetCreate(data_dir, output_dir, use_tokenizer,
multi_lines)
data_creator.create_dataset()
if __name__ == '__main__':
main()
demo/sentiment/sentiment_net.py
浏览文件 @ 84cab2c7
......@@ -47,10 +47,12 @@ def sentiment_data(data_dir=None,
for i, line in enumerate(open(dict_file, 'r')):
word_dict[line.split('\t')[0]] = i
define_py_data_sources2(train_list, test_list,
module="dataprovider",
obj="process",
args={'dictionary': word_dict})
define_py_data_sources2(
train_list,
test_list,
module="dataprovider",
obj="process",
args={'dictionary': word_dict})
return dict_dim, class_dim
......@@ -64,8 +66,7 @@ def bidirectional_lstm_net(input_dim,
emb = embedding_layer(input=data, size=emb_dim)
bi_lstm = bidirectional_lstm(input=emb, size=lstm_dim)
dropout = dropout_layer(input=bi_lstm, dropout_rate=0.5)
output = fc_layer(input=dropout, size=class_dim,
act=SoftmaxActivation())
output = fc_layer(input=dropout, size=class_dim, act=SoftmaxActivation())
if not is_predict:
lbl = data_layer("label", 1)
......@@ -109,27 +110,36 @@ def stacked_lstm_net(input_dim,
data = data_layer("word", input_dim)
emb = embedding_layer(input=data, size=emb_dim)
fc1 = fc_layer(input=emb, size=hid_dim, act=linear,
bias_attr=bias_attr)
lstm1 = lstmemory(input=fc1, act=relu, bias_attr=bias_attr,
layer_attr=layer_attr)
fc1 = fc_layer(input=emb, size=hid_dim, act=linear, bias_attr=bias_attr)
lstm1 = lstmemory(
input=fc1, act=relu, bias_attr=bias_attr, layer_attr=layer_attr)
inputs = [fc1, lstm1]
for i in range(2, stacked_num + 1):
fc = fc_layer(input=inputs, size=hid_dim, act=linear,
param_attr=para_attr, bias_attr=bias_attr)
lstm = lstmemory(input=fc, reverse=(i % 2) == 0, act=relu,
bias_attr=bias_attr, layer_attr=layer_attr)
fc = fc_layer(
input=inputs,
size=hid_dim,
act=linear,
param_attr=para_attr,
bias_attr=bias_attr)
lstm = lstmemory(
input=fc,
reverse=(i % 2) == 0,
act=relu,
bias_attr=bias_attr,
layer_attr=layer_attr)
inputs = [fc, lstm]
fc_last = pooling_layer(input=inputs[0], pooling_type=MaxPooling())
lstm_last = pooling_layer(input=inputs[1], pooling_type=MaxPooling())
output = fc_layer(input=[fc_last, lstm_last], size=class_dim,
act=SoftmaxActivation(),
bias_attr=bias_attr, param_attr=para_attr)
output = fc_layer(
input=[fc_last, lstm_last],
size=class_dim,
act=SoftmaxActivation(),
bias_attr=bias_attr,
param_attr=para_attr)
if is_predict:
outputs(output)
else:
outputs(
classification_cost(input=output, label=data_layer('label', 1)))
outputs(classification_cost(input=output, label=data_layer('label', 1)))
demo/sentiment/trainer_config.py
浏览文件 @ 84cab2c7
......@@ -20,20 +20,19 @@ is_test = get_config_arg('is_test', bool, False)
# whether this config is used for prediction
is_predict = get_config_arg('is_predict', bool, False)
data_dir = "./data/pre-imdb"
data_dir = "./data/pre-imdb"
dict_dim, class_dim = sentiment_data(data_dir, is_test, is_predict)
################## Algorithm Config #####################
settings(
batch_size=128,
learning_rate=2e-3,
learning_method=AdamOptimizer(),
regularization=L2Regularization(8e-4),
gradient_clipping_threshold=25
)
batch_size=128,
learning_rate=2e-3,
learning_method=AdamOptimizer(),
regularization=L2Regularization(8e-4),
gradient_clipping_threshold=25)
#################### Network Config ######################
stacked_lstm_net(dict_dim, class_dim=class_dim,
stacked_num=3, is_predict=is_predict)
stacked_lstm_net(
dict_dim, class_dim=class_dim, stacked_num=3, is_predict=is_predict)
# bidirectional_lstm_net(dict_dim, class_dim=class_dim, is_predict=is_predict)
demo/seqToseq/dataprovider.py
浏览文件 @ 84cab2c7
......@@ -30,14 +30,14 @@ def hook(settings, src_dict, trg_dict, file_list, **kwargs):
if settings.job_mode:
settings.trg_dict = trg_dict
settings.slots = [
integer_value_sequence(len(settings.src_dict)),
integer_value_sequence(len(settings.trg_dict)),
integer_value_sequence(len(settings.src_dict)),
integer_value_sequence(len(settings.trg_dict)),
integer_value_sequence(len(settings.trg_dict))
]
settings.logger.info("trg dict len : %d" % (len(settings.trg_dict)))
else:
settings.slots = [
integer_value_sequence(len(settings.src_dict)),
integer_value_sequence(len(settings.src_dict)),
integer_value_sequence(len(open(file_list[0], "r").readlines()))
]
......@@ -62,8 +62,7 @@ def process(settings, file_name):
if settings.job_mode:
trg_seq = line_split[1] # one target sequence
trg_words = trg_seq.split()
trg_ids = [settings.trg_dict.get(w, UNK_IDX)
for w in trg_words]
trg_ids = [settings.trg_dict.get(w, UNK_IDX) for w in trg_words]
# remove sequence whose length > 80 in training mode
if len(src_ids) > 80 or len(trg_ids) > 80:
......
demo/seqToseq/preprocess.py
浏览文件 @ 84cab2c7
......@@ -12,7 +12,6 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Example:
python preprocess.py -i INPUT [-d DICTSIZE] [-m]
......@@ -24,12 +23,13 @@ Options:
-m --mergeDict merge source and target dictionary
"""
import os
import sys
import sys
import string
from optparse import OptionParser
from paddle.utils.preprocess_util import save_list, DatasetCreater
class SeqToSeqDatasetCreater(DatasetCreater):
"""
A class to process data for sequence to sequence application.
......@@ -75,7 +75,7 @@ class SeqToSeqDatasetCreater(DatasetCreater):
if not os.path.exists(output):
os.system(cmd + '> ' + output)
def build_dict(self, file_path, dict_path, dict_size = -1):
def build_dict(self, file_path, dict_path, dict_size=-1):
"""
Create the dictionary for the file, Note that
1. Valid characters include all printable characters
......@@ -99,20 +99,23 @@ class SeqToSeqDatasetCreater(DatasetCreater):
for word in words:
if word not in dictory:
dictory[word] = 1
else:
else:
dictory[word] += 1
output = open(dict_path, "w+")
output.write('<s>\n<e>\n<unk>\n')
count = 3
for key, value in sorted(dictory.items(), key = lambda d:d[1], reverse = True):
for key, value in sorted(
dictory.items(), key=lambda d: d[1], reverse=True):
output.write(key + "\n")
count += 1
if count == dict_size:
break
self.dict_size = count
def create_dataset(self, dict_size = -1, mergeDict = False,
suffixes = ['.src', '.trg']):
def create_dataset(self,
dict_size=-1,
mergeDict=False,
suffixes=['.src', '.trg']):
"""
Create seqToseq dataset
"""
......@@ -135,13 +138,14 @@ class SeqToSeqDatasetCreater(DatasetCreater):
# checkout dataset should be parallel corpora
suffix_len = len(suffixes[0])
for dataset in dataset_list:
file_list = os.listdir(dataset)
if len(file_list) % 2 == 1:
raise RuntimeError("dataset should be parallel corpora")
file_list.sort()
for i in range(0, len(file_list), 2):
if file_list[i][:-suffix_len] != file_list[i + 1][:-suffix_len]:
raise RuntimeError("source and target file name should be equal")
file_list = os.listdir(dataset)
if len(file_list) % 2 == 1:
raise RuntimeError("dataset should be parallel corpora")
file_list.sort()
for i in range(0, len(file_list), 2):
if file_list[i][:-suffix_len] != file_list[i + 1][:-suffix_len]:
raise RuntimeError(
"source and target file name should be equal")
# cat all the files with the same suffix in dataset
for suffix in suffixes:
......@@ -155,16 +159,18 @@ class SeqToSeqDatasetCreater(DatasetCreater):
list = ['train.list', 'test.list', 'gen.list']
for dataset in dataset_list:
outname = os.path.basename(dataset)
self.concat_file(dataset, outname + suffixes[0],
self.concat_file(dataset, outname + suffixes[0],
outname + suffixes[1], dir_list[id], outname)
save_list([os.path.join(dir_list[id], outname)],
save_list([os.path.join(dir_list[id], outname)],
os.path.join(self.output_path, list[id]))
id += 1
# build dictionary for train data
dict = ['src.dict', 'trg.dict']
dict_path = [os.path.join(self.output_path, dict[0]),
os.path.join(self.output_path, dict[1])]
dict_path = [
os.path.join(self.output_path, dict[0]),
os.path.join(self.output_path, dict[1])
]
if mergeDict:
outname = os.path.join(train_dir, train_dataset.split('/')[-1])
print 'build src dictionary for train data'
......@@ -173,22 +179,30 @@ class SeqToSeqDatasetCreater(DatasetCreater):
os.system('cp ' + dict_path[0] + ' ' + dict_path[1])
else:
outname = os.path.join(train_dataset, self.train_dir_name)
for id in range(0,2):
for id in range(0, 2):
suffix = suffixes[id]
print 'build ' + suffix[1:] + ' dictionary for train data'
self.build_dict(outname + suffix, dict_path[id], dict_size)
print 'dictionary size is', self.dict_size
def main():
usage = "usage: \n" \
"python %prog -i INPUT [-d DICTSIZE] [-m]"
parser = OptionParser(usage)
parser.add_option("-i", action="store", dest="input",
help="input original dataset path")
parser.add_option("-d", action="store", dest="dictsize",
help="specified word count of dictionary")
parser.add_option("-m", "--mergeDict", action="store_true", dest="mergeDict",
help="merge source and target dictionary")
parser.add_option(
"-i", action="store", dest="input", help="input original dataset path")
parser.add_option(
"-d",
action="store",
dest="dictsize",
help="specified word count of dictionary")
parser.add_option(
"-m",
"--mergeDict",
action="store_true",
dest="mergeDict",
help="merge source and target dictionary")
(options, args) = parser.parse_args()
if options.input[-1] == os.path.sep:
options.input = options.input[:-1]
......@@ -200,5 +214,6 @@ def main():
data_creator = SeqToSeqDatasetCreater(options.input, output_path)
data_creator.create_dataset(dictsize, options.mergeDict)
if __name__ == "__main__":
main();
main()
demo/seqToseq/seqToseq_net.py
浏览文件 @ 84cab2c7
......@@ -50,16 +50,21 @@ def seq_to_seq_data(data_dir,
trg_dict = None
else:
train_list = os.path.join(data_dir, train_list)
test_list = os.path.join(data_dir,test_list)
test_list = os.path.join(data_dir, test_list)
define_py_data_sources2(train_list, test_list,
module = "dataprovider",
obj = "process",
args = {"src_dict": src_dict,
"trg_dict": trg_dict})
define_py_data_sources2(
train_list,
test_list,
module="dataprovider",
obj="process",
args={"src_dict": src_dict,
"trg_dict": trg_dict})
return {"src_dict_path": src_lang_dict, "trg_dict_path": trg_lang_dict,
"gen_result": gen_result}
return {
"src_dict_path": src_lang_dict,
"trg_dict_path": trg_lang_dict,
"gen_result": gen_result
}
def gru_encoder_decoder(data_conf,
......@@ -90,51 +95,55 @@ def gru_encoder_decoder(data_conf,
size=word_vector_dim,
param_attr=ParamAttr(name='_source_language_embedding'))
src_forward = simple_gru(input=src_embedding, size=encoder_size)
src_backward = simple_gru(input=src_embedding,
size=encoder_size,
reverse=True)
src_backward = simple_gru(
input=src_embedding, size=encoder_size, reverse=True)
encoded_vector = concat_layer(input=[src_forward, src_backward])
with mixed_layer(size=decoder_size) as encoded_proj:
encoded_proj += full_matrix_projection(input=encoded_vector)
backward_first = first_seq(input=src_backward)
with mixed_layer(size=decoder_size,
act=TanhActivation(), ) as decoder_boot:
with mixed_layer(
size=decoder_size,
act=TanhActivation(), ) as decoder_boot:
decoder_boot += full_matrix_projection(input=backward_first)
def gru_decoder_with_attention(enc_vec, enc_proj, current_word):
decoder_mem = memory(name='gru_decoder',
size=decoder_size,
boot_layer=decoder_boot)
decoder_mem = memory(
name='gru_decoder', size=decoder_size, boot_layer=decoder_boot)
context = simple_attention(encoded_sequence=enc_vec,
encoded_proj=enc_proj,
decoder_state=decoder_mem, )
context = simple_attention(
encoded_sequence=enc_vec,
encoded_proj=enc_proj,
decoder_state=decoder_mem, )
with mixed_layer(size=decoder_size * 3) as decoder_inputs:
decoder_inputs += full_matrix_projection(input=context)
decoder_inputs += full_matrix_projection(input=current_word)
gru_step = gru_step_layer(name='gru_decoder',
input=decoder_inputs,
output_mem=decoder_mem,
size=decoder_size)
gru_step = gru_step_layer(
name='gru_decoder',
input=decoder_inputs,
output_mem=decoder_mem,
size=decoder_size)
with mixed_layer(size=target_dict_dim,
bias_attr=True,
act=SoftmaxActivation()) as out:
with mixed_layer(
size=target_dict_dim, bias_attr=True,
act=SoftmaxActivation()) as out:
out += full_matrix_projection(input=gru_step)
return out
decoder_group_name = "decoder_group"
group_inputs=[StaticInput(input=encoded_vector,is_seq=True),
StaticInput(input=encoded_proj,is_seq=True)]
group_inputs = [
StaticInput(
input=encoded_vector, is_seq=True), StaticInput(
input=encoded_proj, is_seq=True)
]
if not is_generating:
trg_embedding = embedding_layer(
input=data_layer(name='target_language_word',
size=target_dict_dim),
input=data_layer(
name='target_language_word', size=target_dict_dim),
size=word_vector_dim,
param_attr=ParamAttr(name='_target_language_embedding'))
group_inputs.append(trg_embedding)
......@@ -144,12 +153,12 @@ def gru_encoder_decoder(data_conf,
# while encoded source sequence is accessed to as an unbounded memory.
# Here, the StaticInput defines a read-only memory
# for the recurrent_group.
decoder = recurrent_group(name=decoder_group_name,
step=gru_decoder_with_attention,
input=group_inputs)
decoder = recurrent_group(
name=decoder_group_name,
step=gru_decoder_with_attention,
input=group_inputs)
lbl = data_layer(name='target_language_next_word',
size=target_dict_dim)
lbl = data_layer(name='target_language_next_word', size=target_dict_dim)
cost = classification_cost(input=decoder, label=lbl)
outputs(cost)
else:
......@@ -168,16 +177,19 @@ def gru_encoder_decoder(data_conf,
embedding_size=word_vector_dim)
group_inputs.append(trg_embedding)
beam_gen = beam_search(name=decoder_group_name,
step=gru_decoder_with_attention,
input=group_inputs,
bos_id=0,
eos_id=1,
beam_size=beam_size,
max_length=max_length)
seqtext_printer_evaluator(input=beam_gen,
id_input=data_layer(name="sent_id", size=1),
dict_file=trg_dict_path,
result_file=gen_trans_file)
beam_gen = beam_search(
name=decoder_group_name,
step=gru_decoder_with_attention,
input=group_inputs,
bos_id=0,
eos_id=1,
beam_size=beam_size,
max_length=max_length)
seqtext_printer_evaluator(
input=beam_gen,
id_input=data_layer(
name="sent_id", size=1),
dict_file=trg_dict_path,
result_file=gen_trans_file)
outputs(beam_gen)
demo/sequence_tagging/dataprovider.py
浏览文件 @ 84cab2c7
......@@ -17,8 +17,7 @@ import gzip
import logging
logging.basicConfig(
format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s',
)
format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s', )
logger = logging.getLogger('paddle')
logger.setLevel(logging.INFO)
......@@ -32,59 +31,58 @@ num_original_columns = 3
# [[-1,0], [0,0]] means previous token at column 0 and current token at
# column 0 are combined as one feature.
patterns = [
[[-2,0]],
[[-1,0]],
[[0,0]],
[[1,0]],
[[2,0]],
[[-1,0], [0,0]],
[[0,0], [1,0]],
[[-2,1]],
[[-1,1]],
[[0,1]],
[[1,1]],
[[2,1]],
[[-2,1], [-1,1]],
[[-1,1], [0,1]],
[[0,1], [1,1]],
[[1,1], [2,1]],
[[-2,1], [-1,1], [0,1]],
[[-1,1], [0,1], [1,1]],
[[0,1], [1,1], [2,1]],
[[-2, 0]],
[[-1, 0]],
[[0, 0]],
[[1, 0]],
[[2, 0]],
[[-1, 0], [0, 0]],
[[0, 0], [1, 0]],
[[-2, 1]],
[[-1, 1]],
[[0, 1]],
[[1, 1]],
[[2, 1]],
[[-2, 1], [-1, 1]],
[[-1, 1], [0, 1]],
[[0, 1], [1, 1]],
[[1, 1], [2, 1]],
[[-2, 1], [-1, 1], [0, 1]],
[[-1, 1], [0, 1], [1, 1]],
[[0, 1], [1, 1], [2, 1]],
]
dict_label = {
'B-ADJP': 0,
'I-ADJP': 1,
'B-ADVP': 2,
'I-ADVP': 3,
'B-CONJP': 4,
'I-CONJP': 5,
'B-INTJ': 6,
'I-INTJ': 7,
'B-LST': 8,
'I-LST': 9,
'B-NP': 10,
'I-NP': 11,
'B-PP': 12,
'I-PP': 13,
'B-PRT': 14,
'I-PRT': 15,
'B-SBAR': 16,
'I-SBAR': 17,
'B-UCP': 18,
'I-UCP': 19,
'B-VP': 20,
'I-VP': 21,
'O': 22
'B-ADJP': 0,
'I-ADJP': 1,
'B-ADVP': 2,
'I-ADVP': 3,
'B-CONJP': 4,
'I-CONJP': 5,
'B-INTJ': 6,
'I-INTJ': 7,
'B-LST': 8,
'I-LST': 9,
'B-NP': 10,
'I-NP': 11,
'B-PP': 12,
'I-PP': 13,
'B-PRT': 14,
'I-PRT': 15,
'B-SBAR': 16,
'I-SBAR': 17,
'B-UCP': 18,
'I-UCP': 19,
'B-VP': 20,
'I-VP': 21,
'O': 22
}
def make_features(sequence):
length = len(sequence)
num_features = len(sequence[0])
def get_features(pos):
if pos < 0:
return ['#B%s' % -pos] * num_features
......@@ -94,9 +92,10 @@ def make_features(sequence):
for i in xrange(length):
for pattern in patterns:
fname = '/'.join([get_features(i+pos)[f] for pos, f in pattern])
fname = '/'.join([get_features(i + pos)[f] for pos, f in pattern])
sequence[i].append(fname)
'''
Source file format:
Each line is for one timestep. The features are separated by space.
......@@ -109,6 +108,8 @@ i-th column.
return a list of dict for each column
'''
def create_dictionaries(filename, cutoff, oov_policy):
def add_to_dict(sequence, dicts):
num_features = len(dicts)
......@@ -140,7 +141,6 @@ def create_dictionaries(filename, cutoff, oov_policy):
features = line.split(' ')
sequence.append(features)
for i in xrange(num_features):
dct = dicts[i]
n = 1 if oov_policy[i] == OOV_POLICY_USE else 0
......@@ -151,7 +151,7 @@ def create_dictionaries(filename, cutoff, oov_policy):
else:
dct[k] = n
n += 1
if oov_policy[i] == OOV_POLICY_USE:
# placeholder so that len(dct) will be the number of features
# including OOV
......@@ -187,12 +187,15 @@ def initializer(settings, **xargs):
logger.info("feature size=%s" % dim)
settings.input_types = input_types
'''
if oov_policy[i] == OOV_POLICY_USE, features in i-th column which are not
existed in dicts[i] will be assigned to id 0.
if oov_policy[i] == OOV_POLICY_ERROR, all features in i-th column MUST exist
in dicts[i].
'''
@provider(init_hook=initializer, cache=CacheType.CACHE_PASS_IN_MEM)
def process(settings, filename):
input_file = filename
......@@ -231,7 +234,7 @@ def process(settings, filename):
logger.fatal("Unknown token: %s" % features[i])
else:
vec.ids.append(dim + 0)
dim += len(dicts[i])
sample[-1].append(vec)
return sample
......@@ -255,4 +258,3 @@ def process(settings, filename):
f.close()
logger.info("num_sequences=%s" % num_sequences)
demo/sequence_tagging/linear_crf.py
浏览文件 @ 84cab2c7
......@@ -16,11 +16,11 @@ from paddle.trainer_config_helpers import *
import math
define_py_data_sources2(train_list="data/train.list",
test_list="data/test.list",
module="dataprovider",
obj="process")
define_py_data_sources2(
train_list="data/train.list",
test_list="data/test.list",
module="dataprovider",
obj="process")
batch_size = 1
settings(
......@@ -30,14 +30,15 @@ settings(
average_window=0.5,
learning_rate=1e-1,
learning_rate_decay_a=1e-5,
learning_rate_decay_b=0.25,
)
learning_rate_decay_b=0.25, )
num_label_types = 23
num_label_types=23
def get_simd_size(size):
return int(math.ceil(float(size) / 8)) * 8
# Currently, in order to use sparse_update=True,
# the size has to be aligned.
num_label_types = get_simd_size(num_label_types)
......@@ -45,40 +46,37 @@ num_label_types = get_simd_size(num_label_types)
features = data_layer(name="features", size=76328)
word = data_layer(name="word", size=6778)
pos = data_layer(name="pos", size=44)
chunk = data_layer(name="chunk",
size=num_label_types)
chunk = data_layer(name="chunk", size=num_label_types)
crf_input = fc_layer(
input=features,
size=num_label_types,
act=LinearActivation(),
bias_attr=False,
param_attr=ParamAttr(initial_std=0, sparse_update=True))
param_attr=ParamAttr(
initial_std=0, sparse_update=True))
crf=crf_layer(
crf = crf_layer(
input=crf_input,
label=chunk,
param_attr=ParamAttr(name="crfw", initial_std=0),
)
param_attr=ParamAttr(
name="crfw", initial_std=0), )
crf_decoding=crf_decoding_layer(
crf_decoding = crf_decoding_layer(
size=num_label_types,
input=crf_input,
label=chunk,
param_attr=ParamAttr(name="crfw"),
)
param_attr=ParamAttr(name="crfw"), )
sum_evaluator(
name="error",
input=crf_decoding,
)
input=crf_decoding, )
chunk_evaluator(
name="chunk_f1",
input =[crf_decoding, chunk],
input=[crf_decoding, chunk],
chunk_scheme="IOB",
num_chunk_types=11,
)
num_chunk_types=11, )
inputs(word, pos, chunk, features)
outputs(crf)
demo/sequence_tagging/rnn_crf.py
浏览文件 @ 84cab2c7
......@@ -16,10 +16,11 @@ from paddle.trainer_config_helpers import *
import math
define_py_data_sources2(train_list="data/train.list",
test_list="data/test.list",
module="dataprovider",
obj="process")
define_py_data_sources2(
train_list="data/train.list",
test_list="data/test.list",
module="dataprovider",
obj="process")
batch_size = 16
settings(
......@@ -27,29 +28,27 @@ settings(
batch_size=batch_size,
regularization=L2Regularization(batch_size * 1e-5),
average_window=0.5,
learning_rate = 2e-3,
learning_rate_decay_a = 5e-7,
learning_rate_decay_b = 0.5,
)
learning_rate=2e-3,
learning_rate_decay_a=5e-7,
learning_rate_decay_b=0.5, )
word_dim=128
word_dim = 128
hidden_dim = 128
with_rnn = True
initial_std=1/math.sqrt(hidden_dim)
param_attr=ParamAttr(initial_std=initial_std)
cpu_layer_attr=ExtraLayerAttribute(device=-1)
initial_std = 1 / math.sqrt(hidden_dim)
param_attr = ParamAttr(initial_std=initial_std)
cpu_layer_attr = ExtraLayerAttribute(device=-1)
default_device(0)
num_label_types=23
num_label_types = 23
features = data_layer(name="features", size=76328)
word = data_layer(name="word", size=6778)
pos = data_layer(name="pos", size=44)
chunk = data_layer(name="chunk",
size=num_label_types,
layer_attr=cpu_layer_attr)
chunk = data_layer(
name="chunk", size=num_label_types, layer_attr=cpu_layer_attr)
emb = embedding_layer(
input=word, size=word_dim, param_attr=ParamAttr(initial_std=0))
......@@ -58,73 +57,64 @@ hidden1 = mixed_layer(
size=hidden_dim,
act=STanhActivation(),
bias_attr=True,
input=[full_matrix_projection(emb),
table_projection(pos, param_attr=param_attr)]
)
input=[
full_matrix_projection(emb), table_projection(
pos, param_attr=param_attr)
])
if with_rnn:
rnn1 = recurrent_layer(
act=ReluActivation(),
bias_attr=True,
input=hidden1,
param_attr=ParamAttr(initial_std=0),
)
param_attr=ParamAttr(initial_std=0), )
hidden2 = mixed_layer(
size=hidden_dim,
act=STanhActivation(),
bias_attr=True,
input=[full_matrix_projection(hidden1)
] + ([
full_matrix_projection(rnn1, param_attr=ParamAttr(initial_std=0))
] if with_rnn else []),
)
input=[full_matrix_projection(hidden1)] +
([full_matrix_projection(
rnn1, param_attr=ParamAttr(initial_std=0))] if with_rnn else []), )
if with_rnn:
rnn2=recurrent_layer(
rnn2 = recurrent_layer(
reverse=True,
act=ReluActivation(),
bias_attr=True,
input=hidden2,
param_attr=ParamAttr(initial_std=0),
)
param_attr=ParamAttr(initial_std=0), )
crf_input = mixed_layer(
size=num_label_types,
bias_attr=False,
input=[
full_matrix_projection(hidden2),
] + ([
full_matrix_projection(rnn2, param_attr=ParamAttr(initial_std=0))
] if with_rnn else []),
)
input=[full_matrix_projection(hidden2), ] +
([full_matrix_projection(
rnn2, param_attr=ParamAttr(initial_std=0))] if with_rnn else []), )
crf = crf_layer(
input=crf_input,
label=chunk,
param_attr=ParamAttr(name="crfw", initial_std=0),
layer_attr=cpu_layer_attr,
)
param_attr=ParamAttr(
name="crfw", initial_std=0),
layer_attr=cpu_layer_attr, )
crf_decoding = crf_decoding_layer(
size=num_label_types,
input=crf_input,
label=chunk,
param_attr=ParamAttr(name="crfw"),
layer_attr=cpu_layer_attr,
)
layer_attr=cpu_layer_attr, )
sum_evaluator(
name="error",
input=crf_decoding,
)
input=crf_decoding, )
chunk_evaluator(
name="chunk_f1",
input =[crf_decoding, chunk],
input=[crf_decoding, chunk],
chunk_scheme="IOB",
num_chunk_types=11,
)
num_chunk_types=11, )
inputs(word, pos, chunk, features)
outputs(crf)
doc/algorithm/index.rst 0 → 100644
浏览文件 @ 84cab2c7
Algorithm Tutorial
==================
.. toctree::
:maxdepth: 1
rnn/rnn.rst
doc/algorithm/rnn/rnn.rst
浏览文件 @ 84cab2c7
Recurrent Neural Network Configuration
======================================
RNN Configuration
=================
This tutorial will guide you how to configure recurrent neural network in PaddlePaddle. PaddlePaddle supports highly flexible and efficient recurrent neural network configuration. In this tutorial, you will learn how to:
......
doc/build/contribute_to_paddle.md
浏览文件 @ 84cab2c7
# Contribute to PaddlePaddle
# Contribute Code
We sincerely appreciate your contributions. You can use fork and pull request
workflow to merge your code.
......
doc/build/index.rst
浏览文件 @ 84cab2c7
Build And Install PaddlePaddle
================================
Install and Build
=================
Install PaddlePaddle
----------------------
......@@ -18,11 +18,7 @@ Build from Source
.. warning::
Please use :code:`deb` package or :code:`docker` image to install paddle. The building guide is used for hacking or contributing to PaddlePaddle.
If you want to hack and contribute PaddlePaddle source code, following guides can help you\:
Please use :code:`deb` package or :code:`docker` image to install paddle. The building guide is used for hacking or contributing PaddlePaddle source code.
.. toctree::
:maxdepth: 1
......@@ -30,4 +26,3 @@ If you want to hack and contribute PaddlePaddle source code, following guides ca
build_from_source.md
contribute_to_paddle.md
doc/demo/quick_start/index_en.md
浏览文件 @ 84cab2c7
# Quick Start Tutorial
# Quick Start
This tutorial will teach the basics of deep learning (DL), including how to implement many different models in PaddlePaddle. You will learn how to:
- Prepare data into the standardized format that PaddlePaddle accepts.
......
doc/demo/semantic_role_labeling/semantic_role_labeling.md
浏览文件 @ 84cab2c7
# Semantic Role labeling Tutorial #
Semantic role labeling (SRL) is a form of shallow semantic parsing whose goal is to discover the predicate-argument structure of each predicate in a given input sentence. SRL is useful as an intermediate step in a wide range of natural language processing tasks, such as information extraction. automatic document categorization and question answering. An instance is as following [1]:
[ <sub>A0</sub> He ] [ <sub>AM-MOD</sub> would ][ <sub>AM-NEG</sub> n’t ] [ <sub>V</sub> accept] [ <sub>A1</sub> anything of value ] from [<sub>A2</sub> those he was writing about ].
- V: verb
- A0: acceptor
- A1: thing accepted
- A2: accepted-from
- A3: Attribute
- AM-MOD: modal
- AM-NEG: negation
Given the verb "accept", the chunks in sentence would play certain semantic roles. Here, the label scheme is from Penn Proposition Bank.
To this date, most of the successful SRL systems are built on top of some form of parsing results where pre-defined feature templates over the syntactic structure are used. This tutorial will present an end-to-end system using deep bidirectional long short-term memory (DB-LSTM)[2] for solving the SRL task, which largely outperforms the previous state-of-the-art systems. The system regards SRL task as the sequence labelling problem.
## Data Description
The relevant paper[2] takes the data set in CoNLL-2005&2012 Shared Task for training and testing. Accordingto data license, the demo adopts the test data set of CoNLL-2005, which can be reached on website.
To download and process the original data, user just need to execute the following command:
```bash
cd data
./get_data.sh
```
Several new files appear in the `data `directory as follows.
```bash
conll05st-release：the test data set of CoNll-2005 shared task
test.wsj.words：the Wall Street Journal data sentences
test.wsj.props: the propositional arguments
src.dict：the dictionary of words in sentences
tgt.dict：the labels dictionary
feature: the extracted features from data set
```
## Training
### DB-LSTM
Please refer to the Sentiment Analysis demo to learn more about the long short-term memory unit.
Unlike Bidirectional-LSTM that used in Sentiment Analysis demo, the DB-LSTM adopts another way to stack LSTM layer. First a standard LSTM processes the sequence in forward direction. The input and output of this LSTM layer are taken by the next LSTM layer as input, processed in reversed direction. These two standard LSTM layers compose a pair of LSTM. Then we stack LSTM layers pair after pair to obtain the deep LSTM model.
The following figure shows a temporal expanded 2-layer DB-LSTM network.
<center>
![pic](./network_arch.png)
</center>
### Features
Two input features play an essential role in this pipeline: predicate (pred) and argument (argu). Two other features: predicate context (ctx-p) and region mark (mr) are also adopted. Because a single predicate word can not exactly describe the predicate information, especially when the same words appear more than one times in a sentence. With the predicate context, the ambiguity can be largely eliminated. Similarly, we use region mark m<sub>r</sub> = 1 to denote the argument position if it locates in the predicate context region, or m<sub>r</sub> = 0 if does not. These four simple features are all we need for our SRL system. Features of one sample with context size set to 1 is showed as following[2]:
<center>
![pic](./feature.jpg)
</center>
In this sample, the coresponding labelled sentence is:
[ <sub>A1</sub> A record date ] has [ <sub>AM-NEG</sub> n't ] been [ <sub>V</sub> set ] .
In the demo, we adopt the feature template as above, consists of : `argument`, `predicate`, `ctx-p (p=-1,0,1)`, `mark` and use `B/I/O` scheme to label each argument. These features and labels are stored in `feature` file, and separated by `\t`.
### Data Provider
`dataprovider.py` is the python file to wrap data. `hook()` function is to define the data slots for network. The Six features and label are all IndexSlots.
```
def hook(settings, word_dict, label_dict, **kwargs):
settings.word_dict = word_dict
settings.label_dict = label_dict
#all inputs are integral and sequential type
settings.slots = [
integer_value_sequence(len(word_dict)),
integer_value_sequence(len(word_dict)),
integer_value_sequence(len(word_dict)),
integer_value_sequence(len(word_dict)),
integer_value_sequence(len(word_dict)),
integer_value_sequence(2),
integer_value_sequence(len(label_dict))]
```
The corresponding data iterator is as following:
```
@provider(use_seq=True, init_hook=hook)
def process(obj, file_name):
with open(file_name, 'r') as fdata:
for line in fdata:
sentence, predicate, ctx_n1, ctx_0, ctx_p1, mark, label = line.strip().split('\t')
words = sentence.split()
sen_len = len(words)
word_slot = [obj.word_dict.get(w, UNK_IDX) for w in words]
predicate_slot = [obj.word_dict.get(predicate, UNK_IDX)] * sen_len
ctx_n1_slot = [obj.word_dict.get(ctx_n1, UNK_IDX) ] * sen_len
ctx_0_slot = [obj.word_dict.get(ctx_0, UNK_IDX) ] * sen_len
ctx_p1_slot = [obj.word_dict.get(ctx_p1, UNK_IDX) ] * sen_len
marks = mark.split()
mark_slot = [int(w) for w in marks]
label_list = label.split()
label_slot = [obj.label_dict.get(w) for w in label_list]
yield word_slot, predicate_slot, ctx_n1_slot, ctx_0_slot, ctx_p1_slot, mark_slot, label_slot
```
The `process`function yield 7 lists which are six features and labels.
### Neural Network Config
`db_lstm.py` is the neural network config file to load the dictionaries and define the data provider module and network architecture during the training procedure.
Seven `data_layer` load instances from data provider. Six features are transformed into embedddings respectively, and mixed by `mixed_layer` . Deep bidirectional LSTM layers extract features for the softmax layer. The objective function is cross entropy of labels.
### Run Training
The script for training is `train.sh`, user just need to execute:
```bash
./train.sh
```
The content in `train.sh`:
```
paddle train \
--config=./db_lstm.py \
--save_dir=./output \
--trainer_count=4 \
--log_period=10 \
--num_passes=500 \
--use_gpu=false \
--show_parameter_stats_period=10 \
--test_all_data_in_one_period=1 \
2>&1 | tee 'train.log'
```
- \--config=./db_lstm.py : network config file.
- \--save_di=./output: output path to save models.
- \--trainer_count=4 : set thread number (or GPU count).
- \--log_period=10 : print log every 20 batches.
- \--num_passes=500: set pass number, one pass in PaddlePaddle means training all samples in dataset one time.
- \--use_gpu=false: use CPU to train, set true, if you install GPU version of PaddlePaddle and want to use GPU to train.
- \--show_parameter_stats_period=10: show parameter statistic every 100 batches.
- \--test_all_data_in_one_period=1: test all data in every testing.
After training, the models will be saved in directory `output`.
### Run testing
The script for testing is `test.sh`, user just need to execute:
```bash
./test.sh
```
The main part in `tesh.sh`
```
paddle train \
--config=./db_lstm.py \
--model_list=$model_list \
--job=test \
--config_args=is_test=1 \
```
- \--config=./db_lstm.py: network config file
- \--model_list=$model_list.list: model list file
- \--job=test: indicate the test job
- \--config_args=is_test=1: flag to indicate test
### Run prediction
The script for prediction is `predict.sh`, user just need to execute:
```bash
./predict.sh
```
In `predict.sh`, user should offer the network config file, model path, label file, word dictionary file, feature file
```
python predict.py
-c $config_file
-w $model_path
-l $label_file
-d $dict_file
-i $input_file
```
`predict.py` is the main executable python script, which includes functions: load model, load data, data prediction. The network model will output the probability distribution of labels. In the demo, we take the label with maximum probability as result. User can also implement the beam search or viterbi decoding upon the probability distribution matrix.
After prediction, the result is saved in `predict.res`.
## Reference
[1] Martha Palmer, Dan Gildea, and Paul Kingsbury. The Proposition Bank: An Annotated Corpus of Semantic Roles , Computational Linguistics, 31(1), 2005.
[2] Zhou, Jie, and Wei Xu. "End-to-end learning of semantic role labeling using recurrent neural networks." Proceedings of the Annual Meeting of the Association for Computational Linguistics. 2015.
# Semantic Role labeling Tutorial #
Semantic role labeling (SRL) is a form of shallow semantic parsing whose goal is to discover the predicate-argument structure of each predicate in a given input sentence. SRL is useful as an intermediate step in a wide range of natural language processing tasks, such as information extraction. automatic document categorization and question answering. An instance is as following [1]:
[ <sub>A0</sub> He ] [ <sub>AM-MOD</sub> would ][ <sub>AM-NEG</sub> n’t ] [ <sub>V</sub> accept] [ <sub>A1</sub> anything of value ] from [<sub>A2</sub> those he was writing about ].
- V: verb
- A0: acceptor
- A1: thing accepted
- A2: accepted-from
- A3: Attribute
- AM-MOD: modal
- AM-NEG: negation
Given the verb "accept", the chunks in sentence would play certain semantic roles. Here, the label scheme is from Penn Proposition Bank.
To this date, most of the successful SRL systems are built on top of some form of parsing results where pre-defined feature templates over the syntactic structure are used. This tutorial will present an end-to-end system using deep bidirectional long short-term memory (DB-LSTM)[2] for solving the SRL task, which largely outperforms the previous state-of-the-art systems. The system regards SRL task as the sequence labelling problem.
## Data Description
The relevant paper[2] takes the data set in CoNLL-2005&2012 Shared Task for training and testing. Accordingto data license, the demo adopts the test data set of CoNLL-2005, which can be reached on website.
To download and process the original data, user just need to execute the following command:
```bash
cd data
./get_data.sh
```
Several new files appear in the `data `directory as follows.
```bash
conll05st-release：the test data set of CoNll-2005 shared task
test.wsj.words：the Wall Street Journal data sentences
test.wsj.props: the propositional arguments
src.dict：the dictionary of words in sentences
tgt.dict：the labels dictionary
feature: the extracted features from data set
```
## Training
### DB-LSTM
Please refer to the Sentiment Analysis demo to learn more about the long short-term memory unit.
Unlike Bidirectional-LSTM that used in Sentiment Analysis demo, the DB-LSTM adopts another way to stack LSTM layer. First a standard LSTM processes the sequence in forward direction. The input and output of this LSTM layer are taken by the next LSTM layer as input, processed in reversed direction. These two standard LSTM layers compose a pair of LSTM. Then we stack LSTM layers pair after pair to obtain the deep LSTM model.
The following figure shows a temporal expanded 2-layer DB-LSTM network.
<center>
![pic](./network_arch.png)
</center>
### Features
Two input features play an essential role in this pipeline: predicate (pred) and argument (argu). Two other features: predicate context (ctx-p) and region mark (mr) are also adopted. Because a single predicate word can not exactly describe the predicate information, especially when the same words appear more than one times in a sentence. With the predicate context, the ambiguity can be largely eliminated. Similarly, we use region mark m<sub>r</sub> = 1 to denote the argument position if it locates in the predicate context region, or m<sub>r</sub> = 0 if does not. These four simple features are all we need for our SRL system. Features of one sample with context size set to 1 is showed as following[2]:
<center>
![pic](./feature.jpg)
</center>
In this sample, the coresponding labelled sentence is:
[ <sub>A1</sub> A record date ] has [ <sub>AM-NEG</sub> n't ] been [ <sub>V</sub> set ] .
In the demo, we adopt the feature template as above, consists of : `argument`, `predicate`, `ctx-p (p=-1,0,1)`, `mark` and use `B/I/O` scheme to label each argument. These features and labels are stored in `feature` file, and separated by `\t`.
### Data Provider
`dataprovider.py` is the python file to wrap data. `hook()` function is to define the data slots for network. The Six features and label are all IndexSlots.
```
def hook(settings, word_dict, label_dict, **kwargs):
settings.word_dict = word_dict
settings.label_dict = label_dict
#all inputs are integral and sequential type
settings.slots = [
integer_value_sequence(len(word_dict)),
integer_value_sequence(len(word_dict)),
integer_value_sequence(len(word_dict)),
integer_value_sequence(len(word_dict)),
integer_value_sequence(len(word_dict)),
integer_value_sequence(2),
integer_value_sequence(len(label_dict))]
```
The corresponding data iterator is as following:
```
@provider(use_seq=True, init_hook=hook)
def process(obj, file_name):
with open(file_name, 'r') as fdata:
for line in fdata:
sentence, predicate, ctx_n1, ctx_0, ctx_p1, mark, label = line.strip().split('\t')
words = sentence.split()
sen_len = len(words)
word_slot = [obj.word_dict.get(w, UNK_IDX) for w in words]
predicate_slot = [obj.word_dict.get(predicate, UNK_IDX)] * sen_len
ctx_n1_slot = [obj.word_dict.get(ctx_n1, UNK_IDX) ] * sen_len
ctx_0_slot = [obj.word_dict.get(ctx_0, UNK_IDX) ] * sen_len
ctx_p1_slot = [obj.word_dict.get(ctx_p1, UNK_IDX) ] * sen_len
marks = mark.split()
mark_slot = [int(w) for w in marks]
label_list = label.split()
label_slot = [obj.label_dict.get(w) for w in label_list]
yield word_slot, predicate_slot, ctx_n1_slot, ctx_0_slot, ctx_p1_slot, mark_slot, label_slot
```
The `process`function yield 7 lists which are six features and labels.
### Neural Network Config
`db_lstm.py` is the neural network config file to load the dictionaries and define the data provider module and network architecture during the training procedure.
Seven `data_layer` load instances from data provider. Six features are transformed into embedddings respectively, and mixed by `mixed_layer` . Deep bidirectional LSTM layers extract features for the softmax layer. The objective function is cross entropy of labels.
### Run Training
The script for training is `train.sh`, user just need to execute:
```bash
./train.sh
```
The content in `train.sh`:
```
paddle train \
--config=./db_lstm.py \
--save_dir=./output \
--trainer_count=4 \
--log_period=10 \
--num_passes=500 \
--use_gpu=false \
--show_parameter_stats_period=10 \
--test_all_data_in_one_period=1 \
2>&1 | tee 'train.log'
```
- \--config=./db_lstm.py : network config file.
- \--save_di=./output: output path to save models.
- \--trainer_count=4 : set thread number (or GPU count).
- \--log_period=10 : print log every 20 batches.
- \--num_passes=500: set pass number, one pass in PaddlePaddle means training all samples in dataset one time.
- \--use_gpu=false: use CPU to train, set true, if you install GPU version of PaddlePaddle and want to use GPU to train.
- \--show_parameter_stats_period=10: show parameter statistic every 100 batches.
- \--test_all_data_in_one_period=1: test all data in every testing.
After training, the models will be saved in directory `output`.
### Run testing
The script for testing is `test.sh`, user just need to execute:
```bash
./test.sh
```
The main part in `tesh.sh`
```
paddle train \
--config=./db_lstm.py \
--model_list=$model_list \
--job=test \
--config_args=is_test=1 \
```
- \--config=./db_lstm.py: network config file
- \--model_list=$model_list.list: model list file
- \--job=test: indicate the test job
- \--config_args=is_test=1: flag to indicate test
### Run prediction
The script for prediction is `predict.sh`, user just need to execute:
```bash
./predict.sh
```
In `predict.sh`, user should offer the network config file, model path, label file, word dictionary file, feature file
```
python predict.py
-c $config_file
-w $model_path
-l $label_file
-d $dict_file
-i $input_file
```
`predict.py` is the main executable python script, which includes functions: load model, load data, data prediction. The network model will output the probability distribution of labels. In the demo, we take the label with maximum probability as result. User can also implement the beam search or viterbi decoding upon the probability distribution matrix.
After prediction, the result is saved in `predict.res`.
## Reference
[1] Martha Palmer, Dan Gildea, and Paul Kingsbury. The Proposition Bank: An Annotated Corpus of Semantic Roles , Computational Linguistics, 31(1), 2005.
[2] Zhou, Jie, and Wei Xu. "End-to-end learning of semantic role labeling using recurrent neural networks." Proceedings of the Annual Meeting of the Association for Computational Linguistics. 2015.
doc/dev/index.rst 0 → 100644
浏览文件 @ 84cab2c7
Development Guide
=================
.. toctree::
:maxdepth: 1
layer.md
new_layer/new_layer.rst
../source/index.md
doc/dev/layer.md 0 → 100644
浏览文件 @ 84cab2c7
# Layer Documents
* [Layer Source Code Document](../source/gserver/layers/index.rst)
* [Layer Python API Document](../ui/api/trainer_config_helpers/index.rst)
doc/dev/new_layer/index.rst 已删除 100644 → 0
浏览文件 @ 978a6e5e
Writing New Layers
==================
.. toctree::
:maxdepth: 3
new_layer.rst
doc/dev/new_layer/new_layer.rst
浏览文件 @ 84cab2c7
==================
Writing New Layers
==================
......
doc/index.rst 0 → 100644
浏览文件 @ 84cab2c7
PaddlePaddle Documentation
==========================
.. toctree::
:maxdepth: 1
introduction/index.md
user_guide.rst
dev/index.rst
algorithm/index.rst
doc/introduction/index.md
浏览文件 @ 84cab2c7
......@@ -98,4 +98,3 @@ There, you have recovered the underlying pattern between `X` and `Y` only from o
- <a href="../build/index.html"> Build and Installation </a>
- <a href="../demo/quick_start/index_en.html">Quick Start</a>
- <a href="../demo/index.html">Example and Demo</a>
doc/layer.md 已删除 100644 → 0
浏览文件 @ 978a6e5e
# Layer Documents
* [Layer Source Code Document](source/gserver/layers/index.rst)
* [Layer Python API Document](ui/api/trainer_config_helpers/layers_index.rst)
doc/source/gserver/layers/layer.rst
浏览文件 @ 84cab2c7
......@@ -465,6 +465,11 @@ SumOfSquaresCostLayer
.. doxygenclass:: paddle::SumOfSquaresCostLayer
:members:
SumCostLayer
`````````````````````
.. doxygenclass:: paddle::SumCostLayer
:members:
CosSimLayer
-----------
.. doxygenclass:: paddle::CosSimLayer
......
doc/ui/api/trainer_config_helpers/activations.rst
浏览文件 @ 84cab2c7
===========
Activations
===========
BaseActivation
==============
......@@ -102,4 +106,3 @@ STanhActivation
.. automodule:: paddle.trainer_config_helpers.activations
:members: STanhActivation
:noindex:
doc/ui/api/trainer_config_helpers/activations_index.rst 已删除 100644 → 0
浏览文件 @ 978a6e5e
Activations
===========
.. toctree::
:maxdepth: 3
activations.rst
doc/ui/api/trainer_config_helpers/evaluators.rst
浏览文件 @ 84cab2c7
==========
Evaluators
==========
Base
====
.. automodule:: paddle.trainer_config_helpers.evaluators
......
doc/ui/api/trainer_config_helpers/evaluators_index.rst 已删除 100644 → 0
浏览文件 @ 978a6e5e
Evaluators
==========
.. toctree::
:maxdepth: 3
evaluators.rst
doc/ui/api/trainer_config_helpers/index.md 已删除 100644 → 0
浏览文件 @ 978a6e5e
# Model Config Interface
* [Optimizer](optimizers_index.rst)
* [Data Source](data_sources.rst)
* [Layers](layers_index.rst)
* [Activations](activations_index.rst)
* [Poolings](poolings_index.rst)
* [Networks](networks_index.rst)
* [Evaluators](evaluators_index.rst)
* [Parameter and Extra Layer Attribute](attrs.rst)
doc/ui/api/trainer_config_helpers/index.rst 0 → 100644
浏览文件 @ 84cab2c7
Model Config Interface
======================
.. toctree::
:maxdepth: 1
optimizers.rst
data_sources.rst
layers.rst
activations.rst
poolings.rst
networks.rst
evaluators.rst
attrs.rst
doc/ui/api/trainer_config_helpers/layers.rst
浏览文件 @ 84cab2c7
======
Layers
======
Base
======
......@@ -46,6 +50,12 @@ conv_operator
:members: conv_operator
:noindex:
conv_projection
---------------
.. automodule:: paddle.trainer_config_helpers.layers
:members: conv_projection
:noindex:
conv_shift_layer
------------------
.. automodule:: paddle.trainer_config_helpers.layers
......@@ -71,6 +81,12 @@ img_pool_layer
--------------
.. automodule:: paddle.trainer_config_helpers.layers
:members: img_pool_layer
:noindex:
spp_layer
--------------
.. automodule:: paddle.trainer_config_helpers.layers
:members: spp_layer
:noindex:
maxout_layer
......@@ -175,6 +191,12 @@ embedding_layer
:members: embedding_layer
:noindex:
scaling_projection
------------------
.. automodule:: paddle.trainer_config_helpers.layers
:members: scaling_projection
:noindex:
dotmul_projection
-----------------
.. automodule:: paddle.trainer_config_helpers.layers
......@@ -254,6 +276,12 @@ expand_layer
:members: expand_layer
:noindex:
repeat_layer
------------
.. automodule:: paddle.trainer_config_helpers.layers
:members: repeat_layer
:noindex:
Math Layers
===========
......@@ -401,6 +429,12 @@ hsigmoid
:members: hsigmoid
:noindex:
sum_cost
---------
.. automodule:: paddle.trainer_config_helpers.layers
:members: sum_cost
:noindex:
Check Layer
============
......
doc/ui/api/trainer_config_helpers/layers_index.rst 已删除 100644 → 0
浏览文件 @ 978a6e5e
Layers
======
.. toctree::
:maxdepth: 3
layers.rst
doc/ui/api/trainer_config_helpers/networks.rst
浏览文件 @ 84cab2c7
========
Networks
========
The networks module contains pieces of neural network that combine multiple layers.
NLP
===
......@@ -111,4 +117,3 @@ outputs
.. automodule:: paddle.trainer_config_helpers.networks
:members: outputs
:noindex:
doc/ui/api/trainer_config_helpers/networks_index.rst 已删除 100644 → 0
浏览文件 @ 978a6e5e
Networks
========
The networks module contains pieces of neural network that combine multiple layers.
.. toctree::
:maxdepth: 3
networks.rst
doc/ui/api/trainer_config_helpers/optimizers.rst
浏览文件 @ 84cab2c7
==========
Optimizers
==========
BaseSGDOptimizer
================
.. automodule:: paddle.trainer_config_helpers.optimizers
......@@ -51,4 +55,3 @@ settings
.. automodule:: paddle.trainer_config_helpers.optimizers
:members: settings
:noindex:
doc/ui/api/trainer_config_helpers/optimizers_index.rst 已删除 100644 → 0
浏览文件 @ 978a6e5e
Optimizers
==========
.. toctree::
:maxdepth: 3
optimizers.rst
doc/ui/api/trainer_config_helpers/poolings.rst
浏览文件 @ 84cab2c7
========
Poolings
========
BasePoolingType
===============
.. automodule:: paddle.trainer_config_helpers.poolings
......@@ -27,4 +31,3 @@ SquareRootNPooling
.. automodule:: paddle.trainer_config_helpers.poolings
:members: SquareRootNPooling
:noindex:
doc/ui/api/trainer_config_helpers/poolings_index.rst 已删除 100644 → 0
浏览文件 @ 978a6e5e
Poolings
========
These pooling types are used for sequence input, not for images.
.. toctree::
:maxdepth: 3
poolings.rst
doc/ui/predict/predict_sample.py
浏览文件 @ 84cab2c7
......@@ -16,82 +16,113 @@ from py_paddle import swig_paddle, DataProviderConverter
from paddle.trainer.PyDataProvider2 import dense_vector
from paddle.trainer.config_parser import parse_config
TEST_DATA = [[[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.215686,
0.533333, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.67451,
0.992157, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.070588, 0.886275,
0.992157, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.192157, 0.070588, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0.670588, 0.992157, 0.992157, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.117647, 0.933333, 0.858824, 0.313725, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0.090196, 0.858824, 0.992157, 0.831373, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.141176,
0.992157, 0.992157, 0.611765, 0.054902, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.258824, 0.992157, 0.992157,
0.529412, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.368627, 0.992157, 0.992157, 0.419608, 0.003922, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0.094118, 0.835294, 0.992157, 0.992157, 0.517647, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.603922, 0.992157,
0.992157, 0.992157, 0.603922, 0.545098, 0.043137, 0, 0, 0, 0, 0, 0, 0, 0.447059, 0.992157, 0.992157,
0.956863, 0.062745, 0, 0, 0, 0, 0, 0, 0, 0, 0.011765, 0.666667, 0.992157, 0.992157, 0.992157, 0.992157,
0.992157, 0.745098, 0.137255, 0, 0, 0, 0, 0, 0.152941, 0.866667, 0.992157, 0.992157, 0.521569, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0.070588, 0.992157, 0.992157, 0.992157, 0.803922, 0.352941, 0.745098, 0.992157,
0.945098, 0.317647, 0, 0, 0, 0, 0.580392, 0.992157, 0.992157, 0.764706, 0.043137, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0.070588, 0.992157, 0.992157, 0.776471, 0.043137, 0, 0.007843, 0.27451, 0.882353, 0.941176, 0.176471,
0, 0, 0.180392, 0.898039, 0.992157, 0.992157, 0.313725, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.070588, 0.992157,
0.992157, 0.713725, 0, 0, 0, 0, 0.627451, 0.992157, 0.729412, 0.062745, 0, 0.509804, 0.992157, 0.992157,
0.776471, 0.035294, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.494118, 0.992157, 0.992157, 0.968627, 0.168627, 0, 0,
0, 0.423529, 0.992157, 0.992157, 0.364706, 0, 0.717647, 0.992157, 0.992157, 0.317647, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0.533333, 0.992157, 0.984314, 0.945098, 0.603922, 0, 0, 0, 0.003922, 0.466667, 0.992157,
0.988235, 0.976471, 0.992157, 0.992157, 0.788235, 0.007843, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.686275,
0.882353, 0.364706, 0, 0, 0, 0, 0, 0, 0.098039, 0.588235, 0.992157, 0.992157, 0.992157, 0.980392,
0.305882, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.101961, 0.67451, 0.321569, 0, 0, 0, 0, 0, 0, 0, 0.105882,
0.733333, 0.976471, 0.811765, 0.713725, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.65098, 0.992157,
0.321569, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.25098, 0.007843, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1,
0.94902, 0.219608, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.968627,
0.764706, 0.152941, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.498039,
0.25098, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]], [
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.298039, 0.333333, 0.333333, 0.333333, 0.337255, 0.333333,
0.333333, 0.109804, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.027451, 0.223529, 0.776471,
0.964706, 0.988235, 0.988235, 0.988235, 0.992157, 0.988235, 0.988235, 0.780392, 0.098039, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.14902, 0.698039, 0.988235, 0.992157, 0.988235, 0.901961, 0.87451,
0.568627, 0.882353, 0.976471, 0.988235, 0.988235, 0.501961, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0.188235, 0.647059, 0.988235, 0.988235, 0.745098, 0.439216, 0.098039, 0, 0, 0, 0.572549, 0.988235,
0.988235, 0.988235, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.2, 0.933333, 0.992157, 0.941176,
0.247059, 0, 0, 0, 0, 0, 0, 0.188235, 0.898039, 0.992157, 0.992157, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0.039216, 0.639216, 0.933333, 0.988235, 0.913725, 0.278431, 0, 0, 0, 0, 0, 0, 0, 0.113725, 0.843137,
0.988235, 0.988235, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.235294, 0.988235, 0.992157, 0.988235, 0.815686,
0.07451, 0, 0, 0, 0, 0, 0, 0, 0.333333, 0.988235, 0.988235, 0.552941, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0.211765, 0.878431, 0.988235, 0.992157, 0.701961, 0.329412, 0.109804, 0, 0, 0, 0, 0, 0, 0, 0.698039,
0.988235, 0.913725, 0.145098, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.188235, 0.890196, 0.988235, 0.988235,
0.745098, 0.047059, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.882353, 0.988235, 0.568627, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0.2, 0.933333, 0.992157, 0.992157, 0.992157, 0.447059, 0.294118, 0, 0, 0, 0, 0, 0, 0, 0, 0.447059,
0.992157, 0.768627, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.623529, 0.988235, 0.988235, 0.988235, 0.988235,
0.992157, 0.47451, 0, 0, 0, 0, 0, 0, 0, 0.188235, 0.933333, 0.87451, 0.509804, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0.992157, 0.988235, 0.937255, 0.792157, 0.988235, 0.894118, 0.082353, 0, 0, 0, 0, 0, 0,
0.027451, 0.647059, 0.992157, 0.654902, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.623529, 0.988235, 0.913725,
0.329412, 0.376471, 0.184314, 0, 0, 0, 0, 0, 0, 0.027451, 0.513725, 0.988235, 0.635294, 0.219608, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.196078, 0.929412, 0.988235, 0.988235, 0.741176, 0.309804, 0, 0, 0, 0,
0, 0, 0.529412, 0.988235, 0.678431, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.223529, 0.992157,
0.992157, 1, 0.992157, 0.992157, 0.992157, 0.992157, 1, 0.992157, 0.992157, 0.882353, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.023529, 0.478431, 0.654902, 0.658824, 0.952941, 0.988235, 0.988235,
0.988235, 0.992157, 0.988235, 0.729412, 0.278431, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0.196078, 0.647059, 0.764706, 0.764706, 0.768627, 0.580392, 0.047059, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0]]]
TEST_DATA = [[[
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.215686, 0.533333, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.67451, 0.992157, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0.070588, 0.886275, 0.992157, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.192157,
0.070588, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.670588, 0.992157,
0.992157, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.117647, 0.933333, 0.858824, 0.313725,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.090196, 0.858824, 0.992157, 0.831373, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0.141176, 0.992157, 0.992157, 0.611765, 0.054902, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0.258824, 0.992157, 0.992157, 0.529412, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0.368627, 0.992157, 0.992157, 0.419608, 0.003922, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0.094118, 0.835294, 0.992157, 0.992157, 0.517647, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0.603922, 0.992157, 0.992157, 0.992157, 0.603922,
0.545098, 0.043137, 0, 0, 0, 0, 0, 0, 0, 0.447059, 0.992157, 0.992157,
0.956863, 0.062745, 0, 0, 0, 0, 0, 0, 0, 0, 0.011765, 0.666667, 0.992157,
0.992157, 0.992157, 0.992157, 0.992157, 0.745098, 0.137255, 0, 0, 0, 0, 0,
0.152941, 0.866667, 0.992157, 0.992157, 0.521569, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0.070588, 0.992157, 0.992157, 0.992157, 0.803922, 0.352941, 0.745098,
0.992157, 0.945098, 0.317647, 0, 0, 0, 0, 0.580392, 0.992157, 0.992157,
0.764706, 0.043137, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.070588, 0.992157, 0.992157,
0.776471, 0.043137, 0, 0.007843, 0.27451, 0.882353, 0.941176, 0.176471, 0,
0, 0.180392, 0.898039, 0.992157, 0.992157, 0.313725, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0.070588, 0.992157, 0.992157, 0.713725, 0, 0, 0, 0, 0.627451,
0.992157, 0.729412, 0.062745, 0, 0.509804, 0.992157, 0.992157, 0.776471,
0.035294, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.494118, 0.992157, 0.992157,
0.968627, 0.168627, 0, 0, 0, 0.423529, 0.992157, 0.992157, 0.364706, 0,
0.717647, 0.992157, 0.992157, 0.317647, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0.533333, 0.992157, 0.984314, 0.945098, 0.603922, 0, 0, 0, 0.003922,
0.466667, 0.992157, 0.988235, 0.976471, 0.992157, 0.992157, 0.788235,
0.007843, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.686275, 0.882353, 0.364706, 0,
0, 0, 0, 0, 0, 0.098039, 0.588235, 0.992157, 0.992157, 0.992157, 0.980392,
0.305882, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.101961, 0.67451, 0.321569,
0, 0, 0, 0, 0, 0, 0, 0.105882, 0.733333, 0.976471, 0.811765, 0.713725, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.65098, 0.992157, 0.321569, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0.25098, 0.007843, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1,
0.94902, 0.219608, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0.968627, 0.764706, 0.152941, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.498039, 0.25098, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0
]], [[
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0.298039, 0.333333, 0.333333, 0.333333, 0.337255,
0.333333, 0.333333, 0.109804, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0.027451, 0.223529, 0.776471, 0.964706, 0.988235, 0.988235, 0.988235,
0.992157, 0.988235, 0.988235, 0.780392, 0.098039, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0.14902, 0.698039, 0.988235, 0.992157, 0.988235, 0.901961,
0.87451, 0.568627, 0.882353, 0.976471, 0.988235, 0.988235, 0.501961, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.188235, 0.647059, 0.988235, 0.988235,
0.745098, 0.439216, 0.098039, 0, 0, 0, 0.572549, 0.988235, 0.988235,
0.988235, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.2, 0.933333, 0.992157,
0.941176, 0.247059, 0, 0, 0, 0, 0, 0, 0.188235, 0.898039, 0.992157,
0.992157, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.039216, 0.639216, 0.933333,
0.988235, 0.913725, 0.278431, 0, 0, 0, 0, 0, 0, 0, 0.113725, 0.843137,
0.988235, 0.988235, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.235294, 0.988235,
0.992157, 0.988235, 0.815686, 0.07451, 0, 0, 0, 0, 0, 0, 0, 0.333333,
0.988235, 0.988235, 0.552941, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.211765,
0.878431, 0.988235, 0.992157, 0.701961, 0.329412, 0.109804, 0, 0, 0, 0, 0,
0, 0, 0.698039, 0.988235, 0.913725, 0.145098, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0.188235, 0.890196, 0.988235, 0.988235, 0.745098, 0.047059, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0.882353, 0.988235, 0.568627, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.2,
0.933333, 0.992157, 0.992157, 0.992157, 0.447059, 0.294118, 0, 0, 0, 0, 0,
0, 0, 0, 0.447059, 0.992157, 0.768627, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0.623529, 0.988235, 0.988235, 0.988235, 0.988235, 0.992157, 0.47451, 0, 0,
0, 0, 0, 0, 0, 0.188235, 0.933333, 0.87451, 0.509804, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0.992157, 0.988235, 0.937255, 0.792157, 0.988235, 0.894118,
0.082353, 0, 0, 0, 0, 0, 0, 0.027451, 0.647059, 0.992157, 0.654902, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0.623529, 0.988235, 0.913725, 0.329412, 0.376471,
0.184314, 0, 0, 0, 0, 0, 0, 0.027451, 0.513725, 0.988235, 0.635294,
0.219608, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.196078, 0.929412, 0.988235,
0.988235, 0.741176, 0.309804, 0, 0, 0, 0, 0, 0, 0.529412, 0.988235,
0.678431, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.223529, 0.992157,
0.992157, 1, 0.992157, 0.992157, 0.992157, 0.992157, 1, 0.992157, 0.992157,
0.882353, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.023529,
0.478431, 0.654902, 0.658824, 0.952941, 0.988235, 0.988235, 0.988235,
0.992157, 0.988235, 0.729412, 0.278431, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0.196078, 0.647059, 0.764706, 0.764706, 0.768627,
0.580392, 0.047059, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0
]]]
def main():
conf = parse_config("./mnist_model/trainer_config.py", "")
print conf.data_config.load_data_args
network = swig_paddle.GradientMachine.createFromConfigProto(conf.model_config)
network = swig_paddle.GradientMachine.createFromConfigProto(
conf.model_config)
assert isinstance(network, swig_paddle.GradientMachine) # For code hint.
network.loadParameters("./mnist_model/")
converter = DataProviderConverter([dense_vector(784)])
......
doc/user_guide.rst 0 → 100644
浏览文件 @ 84cab2c7
User Guide
==========
.. toctree::
:maxdepth: 1
demo/quick_start/index_en.md
build/index.rst
build/contribute_to_paddle.md
ui/index.md
ui/api/trainer_config_helpers/index.rst
demo/index.md
cluster/index.md
doc_cn/algorithm/rnn/hierarchical-layer.md
浏览文件 @ 84cab2c7
# 支持双层序列作为输入的Layer
## 概述
在自然语言处理任务中，序列是一种常见的数据类型。一个独立的词语，可以看作是一个非序列输入，或者，我们称之为一个0层的序列；由词语构成的句子，是一个单层序列；若干个句子构成一个段落，是一个双层的序列。
双层序列是一个嵌套的序列，它的每一个元素，又是一个单层的序列。这是一种非常灵活的数据组织方式，帮助我们构造一些复杂的输入信息。
我们可以按照如下层次定义非序列，单层序列，以及双层序列。
+ 0层序列：一个独立的元素，类型可以是PaddlePaddle支持的任意输入数据类型
+ 单层序列：排成一列的多个元素，每个元素是一个0层序列，元素之间的顺序是重要的输入信息
+ 双层序列：排成一列的多个元素，每个元素是一个单层序列，称之为双层序列的一个子序列（subseq），subseq的每个元素是一个0层序列
在 PaddlePaddle中，下面这些Layer能够接受双层序列作为输入，完成相应的计算。
## pooling_layer
pooling_layer的使用示例如下，详细见<a href = "../../../doc/ui/api/trainer_config_helpers/layers.html#pooling-layer">配置API</a>。
```python
seq_pool = pooling_layer(input=layer,
pooling_type=AvgPooling(),
agg_level=AggregateLevel.EACH_SEQUENCE)
```
- `pooling_type` 目前支持两种，分别是：MaxPooling()和AvgPooling()。
- `agg_level=AggregateLevel.TIMESTEP`时（默认值）：
- 作用：双层序列经过运算变成一个0层序列，或单层序列经过运算变成一个0层序列
- 输入：一个双层序列，或一个单层序列
- 输出：一个0层序列，即整个输入序列（单层或双层）的平均值（或最大值）
- `agg_level=AggregateLevel.EACH_SEQUENCE`时：
- 作用：一个双层序列经过运算变成一个单层序列
- 输入：必须是一个双层序列
- 输出：一个单层序列，序列的每个元素是原来双层序列每个subseq元素的平均值（或最大值）
## last_seq 和 first_seq
last_seq的使用示例如下（first_seq类似），详细见<a href = "../../../doc/ui/api/trainer_config_helpers/layers.html#last-seq">配置API</a>。
```python
last = last_seq(input=layer,
agg_level=AggregateLevel.EACH_SEQUENCE)
```
- `agg_level=AggregateLevel.TIMESTEP`时（默认值）：
- 作用：一个双层序列经过运算变成一个0层序列，或一个单层序列经过运算变成一个0层序列
- 输入：一个双层序列或一个单层序列
- 输出：一个0层序列，即整个输入序列（双层或者单层）最后一个，或第一个元素。
- `agg_level=AggregateLevel.EACH_SEQUENCE`时：
- 作用：一个双层序列经过运算变成一个单层序列
- 输入：必须是一个双层序列
- 输出：一个单层序列，其中每个元素是双层序列中每个subseq最后一个（或第一个）元素。
## expand_layer
expand_layer的使用示例如下，详细见<a href = "../../../doc/ui/api/trainer_config_helpers/layers.html#expand-layer">配置API</a>。
```python
expand = expand_layer(input=layer1,
expand_as=layer2,
expand_level=ExpandLevel.FROM_TIMESTEP)
```
- `expand_level=ExpandLevel.FROM_TIMESTEP`时（默认值）：
- 作用：一个0层序列经过运算扩展成一个单层序列，或者一个双层序列
- 输入：layer1必须是一个0层序列，是待扩展的数据；layer2可以是一个单层序列，或者是一个双层序列，提供扩展的长度信息
- 输出：一个单层序列，或一个双层序列，输出序列的类型（双层序列，或单层序列）和序列中含有元素的数目同 layer2一致。若输出是单层序列，单层序列的每个元素（0层序列），都是对layer1元素的拷贝；若输出是双层序列，双层序列每个subseq中每个元素（0层序列），都是对layer1元素的拷贝
- `expand_level=ExpandLevel.FROM_SEQUENCE`时：
- 作用：一个单层序列经过运算扩展成一个双层序列
- 输入：layer1必须是一个单层序列，是待扩展的数据；layer2必须是一个双层序列，提供扩展的长度信息
- 输出：一个双层序列，序列中含有元素的数目同layer2一致。要求单层序列含有元素的数目（0层序列），和双层序列含有subseq 的数目一致。单层序列第i个元素（0层序列），被扩展为一个单层序列，构成了输出双层序列的第i个subseq。
\ No newline at end of file
# 支持双层序列作为输入的Layer
## 概述
在自然语言处理任务中，序列是一种常见的数据类型。一个独立的词语，可以看作是一个非序列输入，或者，我们称之为一个0层的序列；由词语构成的句子，是一个单层序列；若干个句子构成一个段落，是一个双层的序列。
双层序列是一个嵌套的序列，它的每一个元素，又是一个单层的序列。这是一种非常灵活的数据组织方式，帮助我们构造一些复杂的输入信息。
我们可以按照如下层次定义非序列，单层序列，以及双层序列。
+ 0层序列：一个独立的元素，类型可以是PaddlePaddle支持的任意输入数据类型
+ 单层序列：排成一列的多个元素，每个元素是一个0层序列，元素之间的顺序是重要的输入信息
+ 双层序列：排成一列的多个元素，每个元素是一个单层序列，称之为双层序列的一个子序列（subseq），subseq的每个元素是一个0层序列
在 PaddlePaddle中，下面这些Layer能够接受双层序列作为输入，完成相应的计算。
## pooling_layer
pooling_layer的使用示例如下，详细见<a href = "../../../doc/ui/api/trainer_config_helpers/layers.html#pooling-layer">配置API</a>。
```python
seq_pool = pooling_layer(input=layer,
pooling_type=AvgPooling(),
agg_level=AggregateLevel.EACH_SEQUENCE)
```
- `pooling_type` 目前支持两种，分别是：MaxPooling()和AvgPooling()。
- `agg_level=AggregateLevel.TIMESTEP`时（默认值）：
- 作用：双层序列经过运算变成一个0层序列，或单层序列经过运算变成一个0层序列
- 输入：一个双层序列，或一个单层序列
- 输出：一个0层序列，即整个输入序列（单层或双层）的平均值（或最大值）
- `agg_level=AggregateLevel.EACH_SEQUENCE`时：
- 作用：一个双层序列经过运算变成一个单层序列
- 输入：必须是一个双层序列
- 输出：一个单层序列，序列的每个元素是原来双层序列每个subseq元素的平均值（或最大值）
## last_seq 和 first_seq
last_seq的使用示例如下（first_seq类似），详细见<a href = "../../../doc/ui/api/trainer_config_helpers/layers.html#last-seq">配置API</a>。
```python
last = last_seq(input=layer,
agg_level=AggregateLevel.EACH_SEQUENCE)
```
- `agg_level=AggregateLevel.TIMESTEP`时（默认值）：
- 作用：一个双层序列经过运算变成一个0层序列，或一个单层序列经过运算变成一个0层序列
- 输入：一个双层序列或一个单层序列
- 输出：一个0层序列，即整个输入序列（双层或者单层）最后一个，或第一个元素。
- `agg_level=AggregateLevel.EACH_SEQUENCE`时：
- 作用：一个双层序列经过运算变成一个单层序列
- 输入：必须是一个双层序列
- 输出：一个单层序列，其中每个元素是双层序列中每个subseq最后一个（或第一个）元素。
## expand_layer
expand_layer的使用示例如下，详细见<a href = "../../../doc/ui/api/trainer_config_helpers/layers.html#expand-layer">配置API</a>。
```python
expand = expand_layer(input=layer1,
expand_as=layer2,
expand_level=ExpandLevel.FROM_TIMESTEP)
```
- `expand_level=ExpandLevel.FROM_TIMESTEP`时（默认值）：
- 作用：一个0层序列经过运算扩展成一个单层序列，或者一个双层序列
- 输入：layer1必须是一个0层序列，是待扩展的数据；layer2可以是一个单层序列，或者是一个双层序列，提供扩展的长度信息
- 输出：一个单层序列，或一个双层序列，输出序列的类型（双层序列，或单层序列）和序列中含有元素的数目同 layer2一致。若输出是单层序列，单层序列的每个元素（0层序列），都是对layer1元素的拷贝；若输出是双层序列，双层序列每个subseq中每个元素（0层序列），都是对layer1元素的拷贝
- `expand_level=ExpandLevel.FROM_SEQUENCE`时：
- 作用：一个单层序列经过运算扩展成一个双层序列
- 输入：layer1必须是一个单层序列，是待扩展的数据；layer2必须是一个双层序列，提供扩展的长度信息
- 输出：一个双层序列，序列中含有元素的数目同layer2一致。要求单层序列含有元素的数目（0层序列），和双层序列含有subseq 的数目一致。单层序列第i个元素（0层序列），被扩展为一个单层序列，构成了输出双层序列的第i个subseq。
doc_cn/algorithm/rnn/rnn-tutorial.md
浏览文件 @ 84cab2c7
......@@ -93,4 +93,4 @@ memory只能在`recurrent_group`中定义和使用。memory不能独立存在，
使用`beam_search`需要遵循以下约定：
- 单层RNN：从一个word生成下一个word。
- 双层RNN：即把单层RNN生成后的subseq给拼接成一个新的双层seq。从语义上看，也不存在一个subseq直接生成下一个subseq的情况。
\ No newline at end of file
- 双层RNN：即把单层RNN生成后的subseq给拼接成一个新的双层seq。从语义上看，也不存在一个subseq直接生成下一个subseq的情况。
doc_cn/build_and_install/install/paddle_version.txt
浏览文件 @ 84cab2c7
......@@ -8,4 +8,4 @@ PaddlePaddle 0.8.0b1, compiled with
with_gflags: ON
with_metric_learning:
with_timer: OFF
with_predict_sdk:
\ No newline at end of file
with_predict_sdk:
doc_cn/concepts/trainer_config.py
浏览文件 @ 84cab2c7
from paddle.trainer_config_helpers import *
define_py_data_sources2(train_list='train.list',
test_list='test.list',
module='provider',
obj='process')
define_py_data_sources2(
train_list='train.list',
test_list='test.list',
module='provider',
obj='process')
settings(
batch_size=128,
learning_rate=1e-3,
learning_method=AdamOptimizer(),
regularization=L2Regularization(0.5)
)
regularization=L2Regularization(0.5))
img = data_layer(name='pixel', size=28 * 28)
hidden1 = simple_img_conv_pool(input=img, filter_size=3, num_filters=32, pool_size=3,
num_channel=1)
hidden1 = simple_img_conv_pool(
input=img, filter_size=3, num_filters=32, pool_size=3, num_channel=1)
hidden2 = fc_layer(input=hidden1, size=200, act=TanhActivation(),
layer_attr=ExtraAttr(drop_rate=0.5))
hidden2 = fc_layer(
input=hidden1,
size=200,
act=TanhActivation(),
layer_attr=ExtraAttr(drop_rate=0.5))
predict = fc_layer(input=hidden2, size=10, act=SoftmaxActivation())
outputs(classification_cost(input=predict, label=data_layer(name='label', size=10)))
outputs(
classification_cost(
input=predict, label=data_layer(
name='label', size=10)))
doc_cn/demo/index.rst
浏览文件 @ 84cab2c7
......@@ -9,7 +9,7 @@
自然语言处理
''''''''''''
* `情感分析 <../../doc/demo/sentiment_analysis/index.html>`_
* `情感分析 <sentiment_analysis/index.html>`_
* `文本生成 <../../doc/demo/text_generation/index.html>`_
* `词性标注 <../../doc/demo/semantic_role_labeling/index.html>`_
......
doc_cn/demo/sentiment_analysis/index.rst 0 → 100644
浏览文件 @ 84cab2c7
情感分析教程
===========================
.. toctree::
:maxdepth: 3
:glob:
Training Locally <sentiment_analysis.md>
\ No newline at end of file
doc_cn/demo/sentiment_analysis/sentiment_analysis.md 0 → 100644
浏览文件 @ 84cab2c7
# 情感分析教程
情感分析有许多应用场景。 一个基本的应用场景是区分给定文本的褒贬两极性，给定的文本可以是一个文档、句子、或者是一个小的文本片段。 一个简单的例子如：把用户在购物网站、旅游网站、团购网站（亚马逊、天猫、淘宝等）上发表的评论分成正面评论和负面评论两类。
情感分析也常用于基于大量评论和个人博客来监控社会媒体。 例如，研究人员分析了几个关于消费者信心和政治观点的调查，结果发现它们与同时期的Twitter消息中的情绪词频率相关 [1]。 另一个例子是通过分析每日Twitter博客的文本内容来预测股票变动 [2]。
另一方面，抓取产品的用户评论并分析他们的情感，有助于理解用户对不同公司，不同产品，甚至不同竞争对手产品的偏好。
本教程将指导您完成长期短期记忆（LSTM）网络的训练过程，以分类来自[大型电影评论数据集](http://ai.stanford.edu/~amaas/data/sentiment/)（有时称为[互联网电影数据库 (IMDB)](http://ai.stanford.edu/~amaas/papers/wvSent_acl2011.pdf)）的句子的情感 。 此数据集包含电影评论及其相关联的类别标签，即正面和负面。
## 数椐准备
### IMDB 数椐介绍
训练模型之前, 我们需要预处理数椐并构建一个字典。 首先, 你可以使用下面的脚本下载 IMDB 数椐集和[Moses](http://www.statmt.org/moses/)工具, 这是一个基于统计的机器翻译系统. 我们提供了一个数据预处理脚本，它不仅能够处理IMDB数据，还能处理其他用户自定义的数据。 为了使用提前编写的脚本，需要将标记的训练和测试样本移动到另一个路径，这已经在`get_imdb.sh`中完成。
```
cd demo/sentiment/data
./get_imdb.sh
```
如果数椐获取成功，你将在目录```./demo/sentiment/data```中看到下面的文件：
```
aclImdb get_imdb.sh imdb mosesdecoder-master
```
* aclImdb: 从外部网站上下载的原始数椐集。
* imdb: 仅包含训练和测试数椐集。
* mosesdecoder-master: Moses 工具。
IMDB数据集包含25,000个已标注过的高极性电影评论用于训练，25,000个用于测试。负面的评论的得分小于等于4,正面的评论的得大于等于7，总评分10分。 运行完脚本 `./get_imdb.sh`后, 我们可以看到在目录 `aclImdb`中的数椐集的结构如下：
```
imdbEr.txt imdb.vocab README test train
```
* train: 训练数椐集。
* test : 测试数椐集。
* imdb.vocab: 字典文件。
* imdbEr.txt: 字典imdb.vocab中每个切分单词的预期评级。
* README: 数椐说明文档。
测试集和训练集目录包含下面的文件:
```
labeledBow.feat neg pos unsup unsupBow.feat urls_neg.txt urls_pos.txt urls_unsup.txt
```
* pos: 正面评价样本，包含12,500个txt文件，每个文件是一个电影评论。
* neg: 负面评价样本，包含12,500个txt文件，每个文件是一个电影评论。
* unsup: 未标记的评价样本，包含50,000个txt文件。
* urls_xx.txt: 每个评论的网址。
* xxBow.feat: 用于统计词频的Bow模型特征。
### IMDB 数椐准备
在这个例子中，我们只使用已经标注过的训练集和测试集，且默认在训练集上构建字典，而不使用IMDB数椐集中的imdb.vocab做为字典。训练集已经做了随机打乱排序而测试集没有。 Moses 工具中的脚本`tokenizer.perl` 用于切分单单词和标点符号。执行下面的命令就可以预处理数椐。
```
cd demo/sentiment/
./preprocess.sh
```
preprocess.sh:
```
data_dir="./data/imdb"
python preprocess.py -i data_dir
```
* data_dir: 输入数椐所在目录。
* preprocess.py: 预处理脚本。
运行成功后目录`demo/sentiment/data/pre-imdb` 结构如下:
```
dict.txt labels.list test.list test_part_000 train.list train_part_000
```
* test\_part\_000 and train\_part\_000: 所有标记的测试集和训练集， 训练集已经随机打乱。
* train.list and test.list: 训练集和测试集文件列表。
* dict.txt: 利用训练集生成的字典。
* labels.txt: neg 0, pos 1, 含义：标签0表示负面的评论，标签1表示正面的评论。
### 用户自定义数椐预处理
如果你执行其它的用情感分析来分类文本的任务，可以按如下的结构来准备数椐. 我们提供了脚本来构建字典和预处理数椐。所以你只用按下面的结构来组织数椐就行了。
```
dataset
|----train
| |----class1
| | |----text_files
| |----class2
| | |----text_files
| | ...
|----test
| |----class1
| | |----text_files
| |----class2
| | |----text_files
| | ...
```
* dataset: 一级目录。
* train, test: 二级目录。
* class1,class2,...: 三级目录。
* text_files: 文本格式的实例文件。
所有同目录下的文本实例文件都是同级别的。 每个文本文件包含一个或者多个实例，每一行表示一个实例。 为了充分的随机打乱训练集, 在预处理含有多行数椐的文本文件时参数设置稍有不同, 执行`preprocess.sh`脚本时需要加上`-m True`参数。 tokenizer.perl 默认用来切分单记和标点符号，如果你不需要这个操作，在运行`preprocess.sh`时加上`-t False`参数即可。
## 训练模型
在这步任务中,我们使用了循环神经网络（RNN）的 LSTM 架构来训练情感分析模型。 引入LSTM模型主要是为了克服消失梯度的问题。 LSTM网络类似于具有隐藏层的标准循环神经网络, 但是隐藏层中的每个普通节点被一个记忆单元替换。 每个记忆单元包含四个主要的元素: 输入门, 具有自循环连接的神经元，忘记门和输出门。 更多的细节可以在文献中找到[4]。 LSTM架构的最大优点是它可以在长时间间隔内记忆信息，而没有短时记忆的损失。在有新的单词来临的每一个时间步骤内，存储在记忆单元区块的历史信息被更新用来迭代的学习单词以合理的序列程现。
<center>![LSTM](../../../doc/demo/sentiment_analysis/lstm.png)</center>
<center>图表 1. LSTM [3]</center>
情感分析是自然语言理解中最典型的问题之一。 它的目的是预测在一个序列中表达的情感态度。 通常, ，仅仅是一些关键词，如形容词和副词，在预测序列或段落的情感中起主要作用。然而有些评论上下文非常长，例如 IMDB的数椐集。 我们只所以使用LSTM来执行这个任务是因为其改进的设计并且具有门机制。 首先，它能够从词级到具有可变上下文长度的上下文级别来总结表示。 第二，它可以在句子级别利用可扩展的上下文, 而大多数方法只是利用n-gram级别的知识。第三，它直接学习段落表示，而不是组合上下文级别信息。
在本演示中，我们提供两个网络，即双向LSTM和三层堆叠LSTM。
#### 双向LSTM
图2是双向LSTM网络，后面连全连接层和softmax层。
<center>![BiLSTM](../../../doc/demo/sentiment_analysis/bi_lstm.jpg)</center>
<center>图 2. Bidirectional-LSTM </center>
#### Stacked-LSTM
图3是三层LSTM结构。图的底部是word embedding(对文档处理后形成的单词向量)。 接下来，连接三个LSTM隐藏层，并且第二个是反向LSTM。然后提取隐藏LSTM层的所有时间步长的最大词向量作为整个序列的表示。 最后，使用具有softmax激活的全连接前馈层来执行分类任务。 更多内容可查看参考文献 [5]。
<center>![StackedLSTM](../../../doc/demo/sentiment_analysis/stacked_lstm.jpg)</center>
<center>图 3. Stacked-LSTM for sentiment analysis </center>
**配置**
进入`demo/sentiment` 目录 , `trainer_config.py` 是一个配置文件的例子, 其中包含算法和网络配置。第一行从`sentiment_net.py`中导出预定义的网络。
trainer_config.py:
```python
from sentiment_net import *
data_dir = "./data/pre-imdb"
# whether this config is used for test
is_test = get_config_arg('is_test', bool, False)
# whether this config is used for prediction
is_predict = get_config_arg('is_predict', bool, False)
dict_dim, class_dim = sentiment_data(data_dir, is_test, is_predict)
################## Algorithm Config #####################
settings(
batch_size=128,
learning_rate=2e-3,
learning_method=AdamOptimizer(),
regularization=L2Regularization(8e-4),
gradient_clipping_threshold=25
)
#################### Network Config ######################
stacked_lstm_net(dict_dim, class_dim=class_dim,
stacked_num=3, is_predict=is_predict)
#bidirectional_lstm_net(dict_dim, class_dim=class_dim, is_predict=is_predict)
```
* **数椐定义**:
* get\_config\_arg(): 获取通过 `--config_args=xx` 设置的命令行参数。
* 定义训练数椐和测试数椐提供者, 这里使用了PaddlePaddle的Python接口来加载数椐。想了解更多细节可以参考PyDataProvider部分的文档
* **算法配置**:
* 使用随机梯度下降（sgd）算法。
* 使用 adam 优化。
* 设置batch size大小为128。
* 设置平均sgd窗口。
* 设置全局学习率。
* **网络配置**:
* dict_dim: 获取字典维度。
* class_dim: 设置类别数，IMDB有两个标签，即正面评价标签和负面评价标签。
* `stacked_lstm_net`: 预定义网络如图3所示，默认情况下使用此网络
* `bidirectional_lstm_net`: 预定义网络，如图2所示。
**训练**
首先安装PaddlePaddle。 然后使用下面的脚本 `train.sh` 来开启本地的训练。
```
cd demo/sentiment/
./train.sh
```
train.sh:
```
config=trainer_config.py
output=./model_output
paddle train --config=$config \
--save_dir=$output \
--job=train \
--use_gpu=false \
--trainer_count=4 \
--num_passes=10 \
--log_period=20 \
--dot_period=20 \
--show_parameter_stats_period=100 \
--test_all_data_in_one_period=1 \
2>&1 | tee 'train.log'
```
* \--config=$config: 设置网络配置。
* \--save\_dir=$output: 设置输出路径以保存训练完成的模型。
* \--job=train: 设置工作模式为训练。
* \--use\_gpu=false: 使用CPU训练，如果你安装GPU版本的PaddlePaddle，并想使用GPU来训练设置为true。
* \--trainer\_count=4:设置线程数（或GPU个数）。
* \--num\_passes=15: 设置pass，PaddlePaddle中的一个pass意味着对数据集中的所有样本进行一次训练。
* \--log\_period=20: 每20个batch打印一次日志。
* \--show\_parameter\_stats\_period=100: 每100个batch打印一次统计信息。
* \--test\_all_data\_in\_one\_period=1: 每次测试都测试所有数据。
如果运行成功，输出日志保存在路径 `demo/sentiment/train.log`中，模型保存在目录`demo/sentiment/model_output/`中。 输出日志说明如下：
```
Batch=20 samples=2560 AvgCost=0.681644 CurrentCost=0.681644 Eval: classification_error_evaluator=0.36875 CurrentEval: classification_error_evaluator=0.36875
...
Pass=0 Batch=196 samples=25000 AvgCost=0.418964 Eval: classification_error_evaluator=0.1922
Test samples=24999 cost=0.39297 Eval: classification_error_evaluator=0.149406
```
- Batch=xx: 表示训练了xx个Batch。
- samples=xx: 表示训练了xx个样本。。
- AvgCost=xx: 从第0个batch到当前batch的平均损失。
- CurrentCost=xx: 最新log_period个batch处理的当前损失。
- Eval: classification\_error\_evaluator=xx: 表示第0个batch到当前batch的分类错误。
- CurrentEval: classification\_error\_evaluator: 最新log_period个batch的分类错误。
- Pass=0: 通过所有训练集一次称为一遍。 0表示第一次经过训练集。
默认情况下，我们使用`stacked_lstm_net`网络，当传递相同的样本数时，它的收敛速度比`bidirectional_lstm_net`快。如果要使用双向LSTM，只需删除最后一行中的注释并把“stacked_lstm_net”注释掉。
## 测试模型
测试模型是指使用训练出的模型评估已标记的验证集。
```
cd demo/sentiment
./test.sh
```
test.sh:
```bash
function get_best_pass() {
cat $1 | grep -Pzo 'Test .*\n.*pass-.*' | \
sed -r 'N;s/Test.* error=([0-9]+\.[0-9]+).*\n.*pass-([0-9]+)/\1 \2/g' | \
sort | head -n 1
}
log=train.log
LOG=`get_best_pass $log`
LOG=(${LOG})
evaluate_pass="model_output/pass-${LOG[1]}"
echo 'evaluating from pass '$evaluate_pass
model_list=./model.list
touch $model_list | echo $evaluate_pass > $model_list
net_conf=trainer_config.py
paddle train --config=$net_conf \
--model_list=$model_list \
--job=test \
--use_gpu=false \
--trainer_count=4 \
--config_args=is_test=1 \
2>&1 | tee 'test.log'
```
函数`get_best_pass`依据分类错误率获得最佳模型进行测试。 在本示例中，我们默认使用IMDB的测试数据集作为验证。 与训练不同，它需要在这里指定`--job = test`和模型路径，即`--model_list = $model_list`。如果运行成功，日志将保存在“demo / sentiment / test.log”的路径中。例如，在我们的测试中，最好的模型是`model_output / pass-00002`，分类误差是0.115645，如下：
```
Pass=0 samples=24999 AvgCost=0.280471 Eval: classification_error_evaluator=0.115645
```
## 预测
`predict.py`脚本提供了一个预测接口。在使用它之前请安装PaddlePaddle的python api。 预测IMDB的未标记评论的一个实例如下：
```
cd demo/sentiment
./predict.sh
```
predict.sh:
```
#Note the default model is pass-00002, you shold make sure the model path
#exists or change the mode path.
model=model_output/pass-00002/
config=trainer_config.py
label=data/pre-imdb/labels.list
python predict.py \
-n $config\
-w $model \
-b $label \
-d data/pre-imdb/dict.txt \
-i data/aclImdb/test/pos/10007_10.txt
```
* `predict.py`: 预测接口脚本。
* -n $config : 设置网络配置。
* -w $model: 设置模型路径。
* -b $label: 设置标签类别字典，这个字典是整数标签和字符串标签的一个对应。
* -d data/pre-imdb/dict.txt: 设置字典文件。
* -i data/aclImdb/test/pos/10014_7.txt: 设置一个要预测的示例文件。
注意应该确保默认模型路径`model_output / pass-00002`存在或更改为其它模型路径。
本示例的预测结果：
```
Loading parameters from model_output/pass-00002/
./data/aclImdb/test/pos/10014_7.txt: predicting label is pos
```
我们真诚地感谢您的关注，并欢迎您来参与贡献。
## 参考文档
[1] Brendan O'Connor, Ramnath Balasubramanyan, Bryan R. Routledge, and Noah A. Smith. 2010. [From Tweets to Polls: Linking Text Sentiment to Public Opinion Time Series](http://homes.cs.washington.edu/~nasmith/papers/oconnor+balasubramanyan+routledge+smith.icwsm10.pdf). In ICWSM-2010. <br>
[2] Johan Bollen, Huina Mao, Xiaojun Zeng. 2011. [Twitter mood predicts the stock market](http://arxiv.org/abs/1010.3003), Journal of Computational Science.<br>
[3] Alex Graves, Marcus Liwicki, Santiago Fernan- dez, Roman Bertolami, Horst Bunke, and Ju ̈rgen Schmidhuber. 2009. [A novel connectionist system for unconstrained handwriting recognition. IEEE Transactions on Pattern Analysis and Machine In- telligence](http://www.cs.toronto.edu/~graves/tpami_2009.pdf), 31(5):855–868.<br>
[4] Zachary C. Lipton, [A Critical Review of Recurrent Neural Networks for Sequence Learning](http://arxiv.org/abs/1506.00019v1), arXiv:1506.00019. <br>
[5] Jie Zhou and Wei Xu; [End-to-end Learning of Semantic Role Labeling Using Recurrent Neural Networks](http://www.aclweb.org/anthology/P/P15/P15-1109.pdf); ACL-IJCNLP 2015. <br>
doc_cn/faq/reduce_min_pool_size.py
浏览文件 @ 84cab2c7
此差异已折叠。
doc_cn/faq/word2vec_config.py
浏览文件 @ 84cab2c7
此差异已折叠。
doc_cn/faq/word2vec_dataprovider.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/.common_test_util.sh
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/CMakeLists.txt
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/api/PaddleAPIPrivate.h
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/api/__init__.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/api/paddle_ld_flags.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/api/test/CMakeLists.txt
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/api/test/testArguments.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/api/test/testMatrix.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/api/test/testTrain.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/api/test/testTrainConfig.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/api/test/testTrainer.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/api/test/testVector.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/cuda/include/hl_base.h
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/cuda/include/hl_cnn.h
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/cuda/include/hl_matrix.h
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/cuda/src/avx_mathfun.h
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/cuda/src/hl_cuda_cnn.cu
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/cuda/src/hl_cuda_matrix.cu
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/CostLayer.cpp
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/CostLayer.h
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/ExpandConvBaseLayer.cpp 0 → 100644
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/ExpandConvBaseLayer.h 0 → 100644
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/ExpandConvTransLayer.cpp 0 → 100644
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/ExpandConvTransLayer.h 0 → 100644
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/PoolLayer.cpp
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/PoolProjection.cpp 0 → 100644
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/PoolProjection.h 0 → 100644
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/Projection.h
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/ScalingProjection.cpp 0 → 100644
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/SpatialPyramidPoolLayer.cpp 0 → 100644
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/layers/SpatialPyramidPoolLayer.h 0 → 100644
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/tests/CMakeLists.txt
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/tests/__init__.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/tests/sequenceGen.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/gserver/tests/test_ConvTrans.cpp 0 → 100644
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/math/BaseMatrix.cu
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/math/BaseMatrix.h
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/math/CpuSparseMatrix.cpp
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/math/MathFunctions.h
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/math/MathUtils.cpp
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/math/MathUtils.h
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/math/Matrix.cpp
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/math/Matrix.h
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/math/Vector.cpp
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/math/Vector.h
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/parameter/CMakeLists.txt
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/py_paddle/__init__.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/py_paddle/util.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/scripts/CMakeLists.txt
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/scripts/cpplint.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/scripts/docker/Dockerfile.m4
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/scripts/docker/build.sh
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/scripts/docker/generate.sh
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/scripts/travis/common.sh
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/trainer/tests/__init__.py
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/trainer/tests/test.txt
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/trainer/tests/train.txt
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/utils/Util.cpp
浏览文件 @ 84cab2c7
此差异已折叠。
paddle/utils/enable_virtualenv.py
浏览文件 @ 84cab2c7
此差异已折叠。
proto/ModelConfig.proto.m4
浏览文件 @ 84cab2c7
此差异已折叠。
python/paddle/__init__.py
浏览文件 @ 84cab2c7
此差异已折叠。
python/paddle/trainer/__init__.py
浏览文件 @ 84cab2c7
此差异已折叠。
python/paddle/utils/__init__.py
浏览文件 @ 84cab2c7
此差异已折叠。
python/paddle/utils/dump_config.py
浏览文件 @ 84cab2c7
此差异已折叠。
python/paddle/utils/image_util.py
浏览文件 @ 84cab2c7
此差异已折叠。
python/paddle/utils/plotcurve.py
浏览文件 @ 84cab2c7
此差异已折叠。
python/paddle/utils/show_pb.py
浏览文件 @ 84cab2c7
此差异已折叠。
python/paddle/utils/torch2paddle.py
浏览文件 @ 84cab2c7
此差异已折叠。
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或