提交 da5aac99 编写于 作者: Z Zrachel 提交者: GitHub

Merge pull request #24 from schinger/sentiment

Sentiment
data/aclImdb
data/imdb
data/pre-imdb
data/mosesdecoder-master
*.log
model_output
dataprovider_copy_1.py
model.list
*.pyc
.DS_Store
此差异已折叠。
#!/bin/bash
# Copyright (c) 2016 PaddlePaddle Authors. 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.
set -e
set -x
DIR="$( cd "$(dirname "$0")" ; pwd -P )"
cd $DIR
#download the dataset
echo "Downloading aclImdb..."
#http://ai.stanford.edu/%7Eamaas/data/sentiment/
wget http://ai.stanford.edu/%7Eamaas/data/sentiment/aclImdb_v1.tar.gz
echo "Downloading mosesdecoder..."
#https://github.com/moses-smt/mosesdecoder
wget https://github.com/moses-smt/mosesdecoder/archive/master.zip
#extract package
echo "Unzipping..."
tar -zxvf aclImdb_v1.tar.gz
unzip master.zip
#move train and test set to imdb_data directory
#in order to process when traing
mkdir -p imdb/train
mkdir -p imdb/test
cp -r aclImdb/train/pos/ imdb/train/pos
cp -r aclImdb/train/neg/ imdb/train/neg
cp -r aclImdb/test/pos/ imdb/test/pos
cp -r aclImdb/test/neg/ imdb/test/neg
#remove compressed package
rm aclImdb_v1.tar.gz
rm master.zip
echo "Done."
# Copyright (c) 2016 PaddlePaddle Authors. 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.
from paddle.trainer.PyDataProvider2 import *
def hook(settings, dictionary, **kwargs):
settings.word_dict = dictionary
settings.input_types = {
'word': integer_value_sequence(len(settings.word_dict)),
'label': integer_value(2)
}
settings.logger.info('dict len : %d' % (len(settings.word_dict)))
@provider(init_hook=hook)
def process(settings, file_name):
with open(file_name, 'r') as fdata:
for line_count, line in enumerate(fdata):
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
]
yield {'word': word_slot, 'label': label}
# Copyright (c) 2016 PaddlePaddle Authors. 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.
import os, sys
import numpy as np
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):
"""
train_conf: trainer configure.
dict_file: word dictionary file name.
model_dir: directory of model.
"""
self.train_conf = train_conf
self.dict_file = dict_file
self.word_dict = {}
self.dict_dim = self.load_dict()
self.model_dir = model_dir
if model_dir is None:
self.model_dir = os.path.dirname(train_conf)
self.label = None
if label_file is not None:
self.load_label(label_file)
conf = parse_config(train_conf, "is_predict=1")
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)
def load_dict(self):
"""
Load dictionary from self.dict_file.
"""
for line_count, line in enumerate(open(self.dict_file, 'r')):
self.word_dict[line.strip().split('\t')[0]] = line_count
return len(self.word_dict)
def load_label(self, label_file):
"""
Load label.
"""
self.label = {}
for v in open(label_file, 'r'):
self.label[int(v.split('\t')[1])] = v.split('\t')[0]
def get_index(self, data):
"""
transform word into integer index according to the dictionary.
"""
words = data.strip().split()
word_slot = [self.word_dict[w] for w in words if w in self.word_dict]
return word_slot
def batch_predict(self, data_batch):
input = self.converter(data_batch)
output = self.network.forwardTest(input)
prob = output[0]["value"]
labs = np.argsort(-prob)
for idx, lab in enumerate(labs):
if self.label is None:
print("predicting label is %d" % (lab[0]))
else:
print("predicting label is %s" % (self.label[lab[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(
"-c",
"--batch_size",
type="int",
action="store",
dest="batch_size",
default=1,
help="the batch size for prediction")
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
batch_size = options.batch_size
dict_file = options.dict_file
model_path = options.model_path
label = options.label
swig_paddle.initPaddle("--use_gpu=0")
predict = SentimentPrediction(train_conf, dict_file, model_path, label)
batch = []
for line in sys.stdin:
batch.append([predict.get_index(line)])
if len(batch) == batch_size:
predict.batch_predict(batch)
batch = []
if len(batch) > 0:
predict.batch_predict(batch)
if __name__ == '__main__':
main()
#!/bin/bash
# Copyright (c) 2016 PaddlePaddle Authors. 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.
set -e
#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
cat ./data/aclImdb/test/pos/10007_10.txt | python predict.py \
--tconf=$config \
--model=$model \
--label=$label \
--dict=./data/pre-imdb/dict.txt \
--batch_size=1
# Copyright (c) 2016 PaddlePaddle Authors. 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.
import os
import sys
import random
import operator
import numpy as np
from subprocess import Popen, PIPE
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):
"""
Save dictionary into file.
dict: input dictionary.
filename: output file name, string.
is_reverse: True, descending order by value.
False, ascending order by value.
"""
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.close()
def tokenize(sentences):
"""
Use tokenizer.perl to tokenize input sentences.
tokenizer.perl is tool of Moses.
sentences : a list of input sentences.
return: a list of processed text.
"""
dir = './data/mosesdecoder-master/scripts/tokenizer/tokenizer.perl'
tokenizer_cmd = [dir, '-l', 'en', '-q', '-']
assert isinstance(sentences, list)
text = "\n".join(sentences)
tokenizer = Popen(tokenizer_cmd, stdin=PIPE, stdout=PIPE)
tok_text, _ = tokenizer.communicate(text)
toks = tok_text.split('\n')[:-1]
return toks
def read_lines(path):
"""
path: String, file path.
return a list of sequence.
"""
seqs = []
with open(path, 'r') as f:
for line in f.readlines():
line = line.strip()
if len(line):
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):
"""
data_path: string, traing and testing dataset path
output_path: string, output path, store processed dataset
multi_lines: whether a file has multi lines.
In order to shuffle fully, it needs to read all files into
memory, then shuffle them if one file has multi lines.
"""
self.output_path = output_path
self.data_path = data_path
self.train_dir = 'train'
self.test_dir = 'test'
self.train_list = "train.list"
self.test_list = "test.list"
self.label_list = "labels.list"
self.classes_num = 0
self.batch_size = 50000
self.batch_dir = 'batches'
self.dict_file = "dict.txt"
self.dict_with_test = False
self.dict_size = 0
self.word_count = {}
self.tokenizer = use_okenizer
self.overwrite = False
self.multi_lines = multi_lines
self.train_dir = join_path(data_path, self.train_dir)
self.test_dir = join_path(data_path, self.test_dir)
self.train_list = join_path(output_path, self.train_list)
self.test_list = join_path(output_path, self.test_list)
self.label_list = join_path(output_path, self.label_list)
self.dict_file = join_path(output_path, self.dict_file)
def data_list(self, path):
"""
create dataset from path
path: data path
return: data list
"""
label_set = get_label_set_from_dir(path)
data = []
for lab_name in label_set.keys():
file_paths = list_files(join_path(path, lab_name))
for p in file_paths:
data.append({"label" : label_set[lab_name],\
"seq_path": p})
return data, label_set
def create_dict(self, data):
"""
create dict for input data.
data: list, [sequence, sequnce, ...]
"""
for seq in data:
for w in seq.strip().lower().split():
if w not in self.word_count:
self.word_count[w] = 1
else:
self.word_count[w] += 1
def create_dataset(self):
"""
create file batches and dictionary of train data set.
If the self.overwrite is false and train.list already exists in
self.output_path, this function will not create and save file
batches from the data set path.
return: dictionary size, class number.
"""
out_path = self.output_path
if out_path and not os.path.exists(out_path):
os.makedirs(out_path)
# If self.overwrite is false or self.train_list has existed,
# it will not process dataset.
if not (self.overwrite or not os.path.exists(self.train_list)):
print "%s already exists." % self.train_list
return
# 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)
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)
print "processing test set..."
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.
save_dict(train_lab_set, self.label_list, False)
self.classes_num = len(train_lab_set.keys())
# save dictionary.
save_dict(self.word_count, self.dict_file, True)
self.dict_size = len(self.word_count)
def save_data(self,
data,
prefix="",
batch_size=50000,
is_shuffle=False,
build_dict=False):
"""
Create batches for a Dataset object.
data: the Dataset object to process.
prefix: the prefix of each batch.
batch_size: number of data in each batch.
build_dict: whether to build dictionary for data
return: list of batch names
"""
if is_shuffle and self.multi_lines:
return self.save_data_multi_lines(data, prefix, batch_size,
build_dict)
if is_shuffle:
random.shuffle(data)
num_batches = int(math.ceil(len(data) / float(batch_size)))
batch_names = []
for i in range(num_batches):
batch_name = join_path(self.output_path,
"%s_part_%03d" % (prefix, i))
begin = i * batch_size
end = min((i + 1) * batch_size, len(data))
# read a batch of data
label_list, data_list = self.get_data_list(begin, end, data)
if build_dict:
self.create_dict(data_list)
self.save_file(label_list, data_list, batch_name)
batch_names.append(batch_name)
return batch_names
def get_data_list(self, begin, end, data):
"""
begin: int, begining index of data.
end: int, ending index of data.
data: a list of {"seq_path": seqquence path, "label": label index}
return a list of label and a list of sequence.
"""
label_list = []
data_list = []
for j in range(begin, end):
seqs = read_lines(data[j]["seq_path"])
lab = int(data[j]["label"])
#File may have multiple lines.
for seq in seqs:
data_list.append(seq)
label_list.append(lab)
if self.tokenizer:
data_list = tokenize(data_list)
return label_list, data_list
def save_data_multi_lines(self,
data,
prefix="",
batch_size=50000,
build_dict=False):
"""
In order to shuffle fully, there is no need to load all data if
each file only contains one sample, it only needs to shuffle list
of file name. But one file contains multi lines, each line is one
sample. It needs to read all data into memory to shuffle fully.
This interface is mainly for data containning multi lines in each
file, which consumes more memory if there is a great mount of data.
data: the Dataset object to process.
prefix: the prefix of each batch.
batch_size: number of data in each batch.
build_dict: whether to build dictionary for data
return: list of batch names
"""
assert self.multi_lines
label_list = []
data_list = []
# read all data
label_list, data_list = self.get_data_list(0, len(data), data)
if build_dict:
self.create_dict(data_list)
length = len(label_list)
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))
begin = i * batch_size
end = min((i + 1) * batch_size, length)
sub_label = [label_list[perm_list[i]] for i in range(begin, end)]
sub_data = [data_list[perm_list[i]] for i in range(begin, end)]
self.save_file(sub_label, sub_data, batch_name)
batch_names.append(batch_name)
return batch_names
def save_file(self, label_list, data_list, filename):
"""
Save data into file.
label_list: a list of int value.
data_list: a list of sequnece.
filename: output file name.
"""
f = open(filename, 'w')
print "saving file: %s" % filename
for lab, seq in zip(label_list, data_list):
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("-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
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.create_dataset()
if __name__ == '__main__':
main()
#!/bin/bash
# Copyright (c) 2016 PaddlePaddle Authors. 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.
set -e
function get_best_pass() {
cat $1 | grep -Pzo 'Test .*\n.*pass-.*' | \
sed -r 'N;s/Test.* classification_error_evaluator=([0-9]+\.[0-9]+).*\n.*pass-([0-9]+)/\1 \2/g' |\
sort -n | 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'
#!/bin/bash
# Copyright (c) 2016 PaddlePaddle Authors. 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.
set -e
paddle train --config=trainer_config.py \
--save_dir=./model_output \
--job=train \
--use_gpu=false \
--trainer_count=4 \
--num_passes=10 \
--log_period=10 \
--dot_period=20 \
--show_parameter_stats_period=100 \
--test_all_data_in_one_period=1 \
2>&1 | tee 'train.log'
# Copyright (c) 2016 PaddlePaddle Authors. 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.
from os.path import join as join_path
from paddle.trainer_config_helpers import *
# 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)
data_dir = "./data/pre-imdb"
train_list = "train.list"
test_list = "test.list"
dict_file = "dict.txt"
dict_dim = len(open(join_path(data_dir, "dict.txt")).readlines())
class_dim = len(open(join_path(data_dir, 'labels.list')).readlines())
if not is_predict:
train_list = join_path(data_dir, train_list)
test_list = join_path(data_dir, test_list)
dict_file = join_path(data_dir, dict_file)
train_list = train_list if not is_test else None
word_dict = dict()
with open(dict_file, 'r') as f:
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})
################## Algorithm Config #####################
settings(
batch_size=128,
learning_rate=2e-3,
learning_method=AdamOptimizer(),
average_window=0.5,
regularization=L2Regularization(8e-4),
gradient_clipping_threshold=25)
#################### Network Config ######################
def convolution_net(input_dim,
class_dim=2,
emb_dim=128,
hid_dim=128,
is_predict=False):
data = data_layer("word", input_dim)
emb = embedding_layer(input=data, size=emb_dim)
conv_3 = sequence_conv_pool(input=emb, context_len=3, hidden_size=hid_dim)
conv_4 = sequence_conv_pool(input=emb, context_len=4, hidden_size=hid_dim)
output = fc_layer(
input=[conv_3, conv_4], size=class_dim, act=SoftmaxActivation())
if not is_predict:
lbl = data_layer("label", 1)
outputs(classification_cost(input=output, label=lbl))
else:
outputs(output)
def stacked_lstm_net(input_dim,
class_dim=2,
emb_dim=128,
hid_dim=512,
stacked_num=3,
is_predict=False):
"""
A Wrapper for sentiment classification task.
This network uses bi-directional recurrent network,
consisting three LSTM layers. This configure is referred to
the paper as following url, but use fewer layrs.
http://www.aclweb.org/anthology/P15-1109
input_dim: here is word dictionary dimension.
class_dim: number of categories.
emb_dim: dimension of word embedding.
hid_dim: dimension of hidden layer.
stacked_num: number of stacked lstm-hidden layer.
is_predict: is predicting or not.
Some layers is not needed in network when predicting.
"""
assert stacked_num % 2 == 1
layer_attr = ExtraLayerAttribute(drop_rate=0.5)
fc_para_attr = ParameterAttribute(learning_rate=1e-3)
lstm_para_attr = ParameterAttribute(initial_std=0., learning_rate=1.)
para_attr = [fc_para_attr, lstm_para_attr]
bias_attr = ParameterAttribute(initial_std=0., l2_rate=0.)
relu = ReluActivation()
linear = LinearActivation()
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)
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)
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)
if is_predict:
outputs(output)
else:
outputs(classification_cost(input=output, label=data_layer('label', 1)))
stacked_lstm_net(
dict_dim, class_dim=class_dim, stacked_num=3, is_predict=is_predict)
# convolution_net(dict_dim, class_dim=class_dim, is_predict=is_predict)
Markdown is supported
0% .
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