Skip to content

M
models

PaddlePaddle / models
1 年多 前同步成功

通知 222
Star 6828
Fork 2962
M
models
未验证 提交 1cb6a643 编写于 作者: L liu zhengxi 提交者： GitHub
浏览文件
操作

update api to 1.8 and update readme (#4609) 

* update api to 1.8 for transformer and similarity_net, test=develop

* update readme, test=develop
上级 35db3d17
展开全部 显示空白变更内容
内联 并排
Showing with 472 addition and 435 deletion
PaddleNLP/machine_translation/transformer/README.md
浏览文件 @ 1cb6a643
...@@ -32,7 +32,7 @@ ...@@ -32,7 +32,7 @@
1. paddle安装 1. paddle安装
本项目依赖于 PaddlePaddle 1.6及以上版本或适当的develop版本，请参考 [安装指南](http://www.paddlepaddle.org/#quick-start) 进行安装 本项目依赖于 PaddlePaddle 1.8及以上版本或适当的develop版本，请参考 [安装指南](https://www.paddlepaddle.org.cn/install/quick) 进行安装
2. 下载代码 2. 下载代码
...@@ -44,7 +44,7 @@ ...@@ -44,7 +44,7 @@
3. 环境依赖 3. 环境依赖
请参考PaddlePaddle[安装说明](https://www.paddlepaddle.org.cn/documentation/docs/zh/1.5/beginners_guide/install/index_cn.html)部分的内容 请参考PaddlePaddle[安装说明](https://www.paddlepaddle.org.cn/documentation/docs/zh/install/index_cn.html)部分的内容
### 数据准备 ### 数据准备
......
PaddleNLP/machine_translation/transformer/transformer.py
浏览文件 @ 1cb6a643
...@@ -752,18 +752,17 @@ def fast_decode(model_input, src_vocab_size, trg_vocab_size, max_in_len, ...@@ -752,18 +752,17 @@ def fast_decode(model_input, src_vocab_size, trg_vocab_size, max_in_len,
# caches contains states of history steps in decoder self-attention # caches contains states of history steps in decoder self-attention
# and static encoder output projections in encoder-decoder attention # and static encoder output projections in encoder-decoder attention
# to reduce redundant computation. # to reduce redundant computation.
batch_size = layers.shape(start_tokens)[0]
caches = [ caches = [
{ {
"k": # for self attention "k": # for self attention
layers.fill_constant_batch_size_like( layers.fill_constant(
input=start_tokens, shape=[batch_size, n_head, 0, d_key],
shape=[-1, n_head, 0, d_key],
dtype=enc_output.dtype, dtype=enc_output.dtype,
value=0), value=0),
"v": # for self attention "v": # for self attention
layers.fill_constant_batch_size_like( layers.fill_constant(
input=start_tokens, shape=[batch_size, n_head, 0, d_value],
shape=[-1, n_head, 0, d_value],
dtype=enc_output.dtype, dtype=enc_output.dtype,
value=0), value=0),
"static_k": # for encoder-decoder attention "static_k": # for encoder-decoder attention
...@@ -792,12 +791,10 @@ def fast_decode(model_input, src_vocab_size, trg_vocab_size, max_in_len, ...@@ -792,12 +791,10 @@ def fast_decode(model_input, src_vocab_size, trg_vocab_size, max_in_len,
lambda x: layers.gather(x, index=gather_idx), caches) lambda x: layers.gather(x, index=gather_idx), caches)
pre_src_attn_bias = layers.gather( pre_src_attn_bias = layers.gather(
trg_src_attn_bias, index=gather_idx) trg_src_attn_bias, index=gather_idx)
bias_batch_size = layers.shape(pre_src_attn_bias)[0]
pre_pos = layers.elementwise_mul( pre_pos = layers.elementwise_mul(
x=layers.fill_constant_batch_size_like( x=layers.fill_constant(
input=pre_src_attn_bias, # cann't use lod tensor here value=1, shape=[bias_batch_size, 1], dtype=pre_ids.dtype),
value=1,
shape=[-1, 1],
dtype=pre_ids.dtype),
y=step_idx, y=step_idx,
axis=0) axis=0)
logits = wrap_decoder( logits = wrap_decoder(
......
PaddleNLP/shared_modules/models/matching/paddle_layers.py
浏览文件 @ 1cb6a643
...@@ -210,7 +210,7 @@ class DataLayer(object): ...@@ -210,7 +210,7 @@ class DataLayer(object):
""" """
operation operation
""" """
data = fluid.layers.data( data = fluid.data(
name=name, shape=shape, dtype=dtype, lod_level=lod_level) name=name, shape=shape, dtype=dtype, lod_level=lod_level)
return data return data
...@@ -383,8 +383,10 @@ class ConstantLayer(object): ...@@ -383,8 +383,10 @@ class ConstantLayer(object):
""" """
operation operation
""" """
constant = fluid.layers.fill_constant_batch_size_like(input, shape, shape = list(shape)
dtype, value) input_shape = fluid.layers.shape(input)
shape[0] = input_shape[0]
constant = fluid.layers.fill_constant(shape, dtype, value)
return constant return constant
......
PaddleNLP/similarity_net/README.md
浏览文件 @ 1cb6a643
...@@ -22,7 +22,7 @@ ...@@ -22,7 +22,7 @@
|UNICOM|联通客服|客服| |UNICOM|联通客服|客服|
## 快速开始 ## 快速开始
#### 版本依赖 #### 版本依赖
本项目依赖于 Paddlepaddle Fluid 1.6，请参考[安装指南](http://www.paddlepaddle.org/#quick-start)进行安装。 本项目依赖于 Paddlepaddle Fluid 1.8，请参考[安装指南](https://www.paddlepaddle.org.cn/install/quick)进行安装。
python版本依赖python 2.7 python版本依赖python 2.7
#### 安装代码 #### 安装代码
......
PaddleNLP/similarity_net/run_classifier.py
浏览文件 @ 1cb6a643
...@@ -47,46 +47,51 @@ from models.model_check import check_version ...@@ -47,46 +47,51 @@ from models.model_check import check_version
from models.model_check import check_cuda from models.model_check import check_cuda
def create_model(args, pyreader_name, is_inference=False, is_pointwise=False): def create_model(args, is_inference=False, is_pointwise=False):
""" """
Create Model for simnet Create Model for simnet
""" """
if is_inference: if is_inference:
inf_pyreader = fluid.layers.py_reader( left = fluid.data(name='left', shape=[None], dtype='int64', lod_level=1)
pos_right = fluid.data(
name='pos_right', shape=[None], dtype='int64', lod_level=1)
inf_loader = fluid.io.DataLoader.from_generator(
capacity=16, capacity=16,
shapes=([-1], [-1]), feed_list=[left, pos_right],
dtypes=('int64', 'int64'), iterable=False,
lod_levels=(1, 1),
name=pyreader_name,
use_double_buffer=False) use_double_buffer=False)
left, pos_right = fluid.layers.read_file(inf_pyreader) return inf_loader, left, pos_right
return inf_pyreader, left, pos_right
else: else:
if is_pointwise: if is_pointwise:
pointwise_pyreader = fluid.layers.py_reader( left = fluid.data(
name='left', shape=[None], dtype='int64', lod_level=1)
right = fluid.data(
name='right', shape=[None], dtype='int64', lod_level=1)
label = fluid.data(name='label', shape=[None], dtype='int64')
pointwise_loader = fluid.io.DataLoader.from_generator(
capacity=16, capacity=16,
shapes=([-1], [-1], [-1]), feed_list=[left, right, label],
dtypes=('int64', 'int64', 'int64'), iterable=False,
lod_levels=(1, 1, 0),
name=pyreader_name,
use_double_buffer=False) use_double_buffer=False)
left, right, label = fluid.layers.read_file(pointwise_pyreader) return pointwise_loader, left, right, label
return pointwise_pyreader, left, right, label
else: else:
pairwise_pyreader = fluid.layers.py_reader( left = fluid.data(
name='left', shape=[None], dtype='int64', lod_level=1)
pos_right = fluid.data(
name='pos_right', shape=[None], dtype='int64', lod_level=1)
neg_right = fluid.data(
name='neg_right', shape=[None], dtype='int64', lod_level=1)
pairwise_loader = fluid.io.DataLoader.from_generator(
capacity=16, capacity=16,
shapes=([-1], [-1], [-1]), feed_list=[left, pos_right, neg_right],
dtypes=('int64', 'int64', 'int64'), iterable=False,
lod_levels=(1, 1, 1),
name=pyreader_name,
use_double_buffer=False) use_double_buffer=False)
left, pos_right, neg_right = fluid.layers.read_file(pairwise_pyreader) return pairwise_loader, left, pos_right, neg_right
return pairwise_pyreader, left, pos_right, neg_right
def train(conf_dict, args): def train(conf_dict, args):
...@@ -131,8 +136,7 @@ def train(conf_dict, args): ...@@ -131,8 +136,7 @@ def train(conf_dict, args):
# Build network # Build network
with fluid.program_guard(train_program, startup_prog): with fluid.program_guard(train_program, startup_prog):
with fluid.unique_name.guard(): with fluid.unique_name.guard():
train_pyreader, left, pos_right, neg_right = create_model( train_loader, left, pos_right, neg_right = create_model(args)
args, pyreader_name='train_reader')
left_feat, pos_score = net.predict(left, pos_right) left_feat, pos_score = net.predict(left, pos_right)
pred = pos_score pred = pos_score
_, neg_score = net.predict(left, neg_right) _, neg_score = net.predict(left, neg_right)
...@@ -147,8 +151,8 @@ def train(conf_dict, args): ...@@ -147,8 +151,8 @@ def train(conf_dict, args):
test_prog = fluid.Program() test_prog = fluid.Program()
with fluid.program_guard(test_prog, startup_prog): with fluid.program_guard(test_prog, startup_prog):
with fluid.unique_name.guard(): with fluid.unique_name.guard():
test_pyreader, left, pos_right = create_model( test_loader, left, pos_right = create_model(
args, pyreader_name='test_reader', is_inference=True) args, is_inference=True)
left_feat, pos_score = net.predict(left, pos_right) left_feat, pos_score = net.predict(left, pos_right)
pred = pos_score pred = pos_score
test_prog = test_prog.clone(for_test=True) test_prog = test_prog.clone(for_test=True)
...@@ -157,8 +161,8 @@ def train(conf_dict, args): ...@@ -157,8 +161,8 @@ def train(conf_dict, args):
# Build network # Build network
with fluid.program_guard(train_program, startup_prog): with fluid.program_guard(train_program, startup_prog):
with fluid.unique_name.guard(): with fluid.unique_name.guard():
train_pyreader, left, right, label = create_model( train_loader, left, right, label = create_model(
args, pyreader_name='train_reader', is_pointwise=True) args, is_pointwise=True)
left_feat, pred = net.predict(left, right) left_feat, pred = net.predict(left, right)
avg_cost = loss.compute(pred, label) avg_cost = loss.compute(pred, label)
avg_cost.persistable = True avg_cost.persistable = True
...@@ -171,15 +175,15 @@ def train(conf_dict, args): ...@@ -171,15 +175,15 @@ def train(conf_dict, args):
test_prog = fluid.Program() test_prog = fluid.Program()
with fluid.program_guard(test_prog, startup_prog): with fluid.program_guard(test_prog, startup_prog):
with fluid.unique_name.guard(): with fluid.unique_name.guard():
test_pyreader, left, right = create_model( test_loader, left, right = create_model(
args, pyreader_name='test_reader', is_inference=True) args, is_inference=True)
left_feat, pred = net.predict(left, right) left_feat, pred = net.predict(left, right)
test_prog = test_prog.clone(for_test=True) test_prog = test_prog.clone(for_test=True)
if args.init_checkpoint is not "": if args.init_checkpoint is not "":
utils.init_checkpoint(exe, args.init_checkpoint, startup_prog) utils.init_checkpoint(exe, args.init_checkpoint, startup_prog)
def valid_and_test(test_program, test_pyreader, get_valid_examples, process, def valid_and_test(test_program, test_loader, get_valid_examples, process,
mode, exe, fetch_list): mode, exe, fetch_list):
""" """
return auc and acc return auc and acc
...@@ -187,15 +191,15 @@ def train(conf_dict, args): ...@@ -187,15 +191,15 @@ def train(conf_dict, args):
# Get Batch Data # Get Batch Data
batch_data = fluid.io.batch( batch_data = fluid.io.batch(
get_valid_examples, args.batch_size, drop_last=False) get_valid_examples, args.batch_size, drop_last=False)
test_pyreader.decorate_paddle_reader(batch_data) test_loader.set_sample_list_generator(batch_data)
test_pyreader.start() test_loader.start()
pred_list = [] pred_list = []
while True: while True:
try: try:
_pred = exe.run(program=test_program, fetch_list=[pred.name]) _pred = exe.run(program=test_program, fetch_list=[pred.name])
pred_list += list(_pred) pred_list += list(_pred)
except fluid.core.EOFException: except fluid.core.EOFException:
test_pyreader.reset() test_loader.reset()
break break
pred_list = np.vstack(pred_list) pred_list = np.vstack(pred_list)
if mode == "test": if mode == "test":
...@@ -233,8 +237,8 @@ def train(conf_dict, args): ...@@ -233,8 +237,8 @@ def train(conf_dict, args):
get_train_examples, buf_size=10000), get_train_examples, buf_size=10000),
args.batch_size, args.batch_size,
drop_last=False) drop_last=False)
train_pyreader.decorate_paddle_reader(train_batch_data) train_loader.set_sample_list_generator(train_batch_data)
train_pyreader.start() train_loader.start()
exe.run(startup_prog) exe.run(startup_prog)
losses = [] losses = []
start_time = time.time() start_time = time.time()
...@@ -248,8 +252,8 @@ def train(conf_dict, args): ...@@ -248,8 +252,8 @@ def train(conf_dict, args):
if args.do_valid and global_step % args.validation_steps == 0: if args.do_valid and global_step % args.validation_steps == 0:
get_valid_examples = simnet_process.get_reader("valid") get_valid_examples = simnet_process.get_reader("valid")
valid_result = valid_and_test( valid_result = valid_and_test(
test_prog, test_pyreader, get_valid_examples, test_prog, test_loader, get_valid_examples, simnet_process,
simnet_process, "valid", exe, [pred.name]) "valid", exe, [pred.name])
if args.compute_accuracy: if args.compute_accuracy:
valid_auc, valid_acc = valid_result valid_auc, valid_acc = valid_result
logging.info( logging.info(
...@@ -281,7 +285,7 @@ def train(conf_dict, args): ...@@ -281,7 +285,7 @@ def train(conf_dict, args):
logging.info("saving infer model in %s" % model_path) logging.info("saving infer model in %s" % model_path)
except fluid.core.EOFException: except fluid.core.EOFException:
train_pyreader.reset() train_loader.reset()
break break
end_time = time.time() end_time = time.time()
#logging.info("epoch: %d, loss: %f, used time: %d sec" % #logging.info("epoch: %d, loss: %f, used time: %d sec" %
...@@ -327,9 +331,8 @@ def train(conf_dict, args): ...@@ -327,9 +331,8 @@ def train(conf_dict, args):
else: else:
# Get Feeder and Reader # Get Feeder and Reader
get_test_examples = simnet_process.get_reader("test") get_test_examples = simnet_process.get_reader("test")
test_result = valid_and_test(test_prog, test_pyreader, test_result = valid_and_test(test_prog, test_loader, get_test_examples,
get_test_examples, simnet_process, "test", simnet_process, "test", exe, [pred.name])
exe, [pred.name])
if args.compute_accuracy: if args.compute_accuracy:
test_auc, test_acc = test_result test_auc, test_acc = test_result
logging.info("AUC of test is %f, Accuracy of test is %f" % logging.info("AUC of test is %f, Accuracy of test is %f" %
...@@ -371,8 +374,8 @@ def test(conf_dict, args): ...@@ -371,8 +374,8 @@ def test(conf_dict, args):
if args.task_mode == "pairwise": if args.task_mode == "pairwise":
with fluid.program_guard(test_prog, startup_prog): with fluid.program_guard(test_prog, startup_prog):
with fluid.unique_name.guard(): with fluid.unique_name.guard():
test_pyreader, left, pos_right = create_model( test_loader, left, pos_right = create_model(
args, pyreader_name='test_reader', is_inference=True) args, is_inference=True)
left_feat, pos_score = net.predict(left, pos_right) left_feat, pos_score = net.predict(left, pos_right)
pred = pos_score pred = pos_score
test_prog = test_prog.clone(for_test=True) test_prog = test_prog.clone(for_test=True)
...@@ -380,8 +383,8 @@ def test(conf_dict, args): ...@@ -380,8 +383,8 @@ def test(conf_dict, args):
else: else:
with fluid.program_guard(test_prog, startup_prog): with fluid.program_guard(test_prog, startup_prog):
with fluid.unique_name.guard(): with fluid.unique_name.guard():
test_pyreader, left, right = create_model( test_loader, left, right = create_model(
args, pyreader_name='test_reader', is_inference=True) args, is_inference=True)
left_feat, pred = net.predict(left, right) left_feat, pred = net.predict(left, right)
test_prog = test_prog.clone(for_test=True) test_prog = test_prog.clone(for_test=True)
...@@ -390,10 +393,10 @@ def test(conf_dict, args): ...@@ -390,10 +393,10 @@ def test(conf_dict, args):
utils.init_checkpoint(exe, args.init_checkpoint, main_program=test_prog) utils.init_checkpoint(exe, args.init_checkpoint, main_program=test_prog)
test_exe = exe test_exe = exe
test_pyreader.decorate_paddle_reader(batch_data) test_loader.set_sample_list_generator(batch_data)
logging.info("start test process ...") logging.info("start test process ...")
test_pyreader.start() test_loader.start()
pred_list = [] pred_list = []
fetch_list = [pred.name] fetch_list = [pred.name]
output = [] output = []
...@@ -412,7 +415,7 @@ def test(conf_dict, args): ...@@ -412,7 +415,7 @@ def test(conf_dict, args):
map(lambda item: str(np.argmax(item)), output[0])) + map(lambda item: str(np.argmax(item)), output[0])) +
"\n") "\n")
except fluid.core.EOFException: except fluid.core.EOFException:
test_pyreader.reset() test_loader.reset()
break break
if args.task_mode == "pairwise": if args.task_mode == "pairwise":
pred_list = np.array(pred_list).reshape((-1, 1)) pred_list = np.array(pred_list).reshape((-1, 1))
...@@ -468,16 +471,16 @@ def infer(conf_dict, args): ...@@ -468,16 +471,16 @@ def infer(conf_dict, args):
if args.task_mode == "pairwise": if args.task_mode == "pairwise":
with fluid.program_guard(test_prog, startup_prog): with fluid.program_guard(test_prog, startup_prog):
with fluid.unique_name.guard(): with fluid.unique_name.guard():
infer_pyreader, left, pos_right = create_model( infer_loader, left, pos_right = create_model(
args, pyreader_name='infer_reader', is_inference=True) args, is_inference=True)
left_feat, pos_score = net.predict(left, pos_right) left_feat, pos_score = net.predict(left, pos_right)
pred = pos_score pred = pos_score
test_prog = test_prog.clone(for_test=True) test_prog = test_prog.clone(for_test=True)
else: else:
with fluid.program_guard(test_prog, startup_prog): with fluid.program_guard(test_prog, startup_prog):
with fluid.unique_name.guard(): with fluid.unique_name.guard():
infer_pyreader, left, right = create_model( infer_loader, left, right = create_model(
args, pyreader_name='infer_reader', is_inference=True) args, is_inference=True)
left_feat, pred = net.predict(left, right) left_feat, pred = net.predict(left, right)
test_prog = test_prog.clone(for_test=True) test_prog = test_prog.clone(for_test=True)
...@@ -486,13 +489,13 @@ def infer(conf_dict, args): ...@@ -486,13 +489,13 @@ def infer(conf_dict, args):
utils.init_checkpoint(exe, args.init_checkpoint, main_program=test_prog) utils.init_checkpoint(exe, args.init_checkpoint, main_program=test_prog)
test_exe = exe test_exe = exe
infer_pyreader.decorate_sample_list_generator(batch_data) infer_loader.set_sample_list_generator(batch_data)
logging.info("start test process ...") logging.info("start test process ...")
preds_list = [] preds_list = []
fetch_list = [pred.name] fetch_list = [pred.name]
output = [] output = []
infer_pyreader.start() infer_loader.start()
while True: while True:
try: try:
output = test_exe.run(program=test_prog, fetch_list=fetch_list) output = test_exe.run(program=test_prog, fetch_list=fetch_list)
...@@ -502,7 +505,7 @@ def infer(conf_dict, args): ...@@ -502,7 +505,7 @@ def infer(conf_dict, args):
else: else:
preds_list += map(lambda item: str(np.argmax(item)), output[0]) preds_list += map(lambda item: str(np.argmax(item)), output[0])
except fluid.core.EOFException: except fluid.core.EOFException:
infer_pyreader.reset() infer_loader.reset()
break break
with io.open(args.infer_result_path, "w", encoding="utf8") as infer_file: with io.open(args.infer_result_path, "w", encoding="utf8") as infer_file:
for _data, _pred in zip(simnet_process.get_infer_data(), preds_list): for _data, _pred in zip(simnet_process.get_infer_data(), preds_list):
......
dygraph/similarity_net/README.md
浏览文件 @ 1cb6a643
...@@ -23,7 +23,7 @@ ...@@ -23,7 +23,7 @@
## 快速开始 ## 快速开始
#### 版本依赖 #### 版本依赖
本项目依赖于 Paddlepaddle Fluid 1.7，请参考[安装指南](http://www.paddlepaddle.org/#quick-start)进行安装。 本项目依赖于 Paddlepaddle Fluid 1.8，请参考[安装指南](https://www.paddlepaddle.org.cn/install/quick)进行安装。
#### 安装代码 #### 安装代码
......
dygraph/similarity_net/model_check.py
浏览文件 @ 1cb6a643
...@@ -33,6 +33,7 @@ def check_cuda(use_cuda, err = \ ...@@ -33,6 +33,7 @@ def check_cuda(use_cuda, err = \
except Exception as e: except Exception as e:
pass pass
def check_version(): def check_version():
""" """
Log error and exit when the installed version of paddlepaddle is Log error and exit when the installed version of paddlepaddle is
...@@ -43,7 +44,7 @@ def check_version(): ...@@ -43,7 +44,7 @@ def check_version():
"Please make sure the version is good with your code." \ "Please make sure the version is good with your code." \
try: try:
fluid.require_version('1.6.0') fluid.require_version('1.8.0')
except Exception as e: except Exception as e:
print(err) print(err)
sys.exit(1) sys.exit(1)
...@@ -59,7 +60,7 @@ def check_version(): ...@@ -59,7 +60,7 @@ def check_version():
"Please make sure the version is good with your code." \ "Please make sure the version is good with your code." \
try: try:
fluid.require_version('1.6.0') fluid.require_version('1.8.0')
except Exception as e: except Exception as e:
print(err) print(err)
sys.exit(1) sys.exit(1)
......
dygraph/similarity_net/nets/paddle_layers.py
浏览文件 @ 1cb6a643
...@@ -30,6 +30,7 @@ import paddle.fluid.layers.utils as utils ...@@ -30,6 +30,7 @@ import paddle.fluid.layers.utils as utils
from paddle.fluid.dygraph import Embedding, Conv2D, GRUUnit, Layer, to_variable from paddle.fluid.dygraph import Embedding, Conv2D, GRUUnit, Layer, to_variable
from paddle.fluid.layers.utils import map_structure, flatten, pack_sequence_as from paddle.fluid.layers.utils import map_structure, flatten, pack_sequence_as
class EmbeddingLayer(object): class EmbeddingLayer(object):
""" """
Embedding Layer class Embedding Layer class
...@@ -52,12 +53,12 @@ class EmbeddingLayer(object): ...@@ -52,12 +53,12 @@ class EmbeddingLayer(object):
size=[self.dict_size, self.emb_dim], size=[self.dict_size, self.emb_dim],
is_sparse=True, is_sparse=True,
padding_idx=self.padding_idx, padding_idx=self.padding_idx,
param_attr=attr.ParamAttr(name=self.name, initializer=fluid.initializer.Xavier())) param_attr=attr.ParamAttr(
name=self.name, initializer=fluid.initializer.Xavier()))
return emb return emb
class FCLayer(object): class FCLayer(object):
""" """
Fully Connect Layer class Fully Connect Layer class
...@@ -117,11 +118,13 @@ class DynamicLSTMLayer(object): ...@@ -117,11 +118,13 @@ class DynamicLSTMLayer(object):
self.lstm_dim = lstm_dim self.lstm_dim = lstm_dim
self.name = name self.name = name
self.is_reverse = is_reverse self.is_reverse = is_reverse
def ops(self): def ops(self):
""" """
operation operation
""" """
lstm_cell = BasicLSTMUnit(hidden_size=self.lstm_dim, input_size=self.lstm_dim*4) lstm_cell = BasicLSTMUnit(
hidden_size=self.lstm_dim, input_size=self.lstm_dim * 4)
lstm = RNN(cell=lstm_cell, time_major=True, is_reverse=self.is_reverse) lstm = RNN(cell=lstm_cell, time_major=True, is_reverse=self.is_reverse)
return lstm return lstm
...@@ -141,7 +144,7 @@ class DataLayer(object): ...@@ -141,7 +144,7 @@ class DataLayer(object):
""" """
operation operation
""" """
data = fluid.layers.data( data = fluid.data(
name=name, shape=shape, dtype=dtype, lod_level=lod_level) name=name, shape=shape, dtype=dtype, lod_level=lod_level)
return data return data
...@@ -314,8 +317,10 @@ class ConstantLayer(object): ...@@ -314,8 +317,10 @@ class ConstantLayer(object):
""" """
operation operation
""" """
constant = fluid.layers.fill_constant_batch_size_like(input, shape, shape = list(shape)
dtype, value) input_shape = fluid.layers.shape(input)
shape[0] = input_shape[0]
constant = fluid.layers.fill_constant(shape, dtype, value)
return constant return constant
...@@ -358,14 +363,10 @@ class SoftsignLayer(object): ...@@ -358,14 +363,10 @@ class SoftsignLayer(object):
class SimpleConvPool(Layer): class SimpleConvPool(Layer):
def __init__(self, def __init__(self, num_channels, num_filters, filter_size, use_cudnn=False):
num_channels,
num_filters,
filter_size,
use_cudnn=False
):
super(SimpleConvPool, self).__init__() super(SimpleConvPool, self).__init__()
self._conv2d = Conv2D(num_channels = num_channels, self._conv2d = Conv2D(
num_channels=num_channels,
num_filters=num_filters, num_filters=num_filters,
filter_size=filter_size, filter_size=filter_size,
padding=[1, 1], padding=[1, 1],
...@@ -378,6 +379,7 @@ class SimpleConvPool(Layer): ...@@ -378,6 +379,7 @@ class SimpleConvPool(Layer):
x = fluid.layers.reshape(x, shape=[x.shape[0], -1]) x = fluid.layers.reshape(x, shape=[x.shape[0], -1])
return x return x
class FC(Layer): class FC(Layer):
""" """
This interface is used to construct a callable object of the ``FC`` class. This interface is used to construct a callable object of the ``FC`` class.
...@@ -580,7 +582,7 @@ class DynamicGRU(Layer): ...@@ -580,7 +582,7 @@ class DynamicGRU(Layer):
gate_activation='sigmoid', gate_activation='sigmoid',
candidate_activation='tanh', candidate_activation='tanh',
origin_mode=False, origin_mode=False,
init_size = None): init_size=None):
super(DynamicGRU, self).__init__() super(DynamicGRU, self).__init__()
self.gru_unit = GRUUnit( self.gru_unit = GRUUnit(
size * 3, size * 3,
...@@ -591,16 +593,19 @@ class DynamicGRU(Layer): ...@@ -591,16 +593,19 @@ class DynamicGRU(Layer):
origin_mode=origin_mode) origin_mode=origin_mode)
self.size = size self.size = size
self.is_reverse = is_reverse self.is_reverse = is_reverse
def forward(self, inputs, h_0): def forward(self, inputs, h_0):
hidden = h_0 hidden = h_0
res = [] res = []
for i in range(inputs.shape[1]): for i in range(inputs.shape[1]):
if self.is_reverse: if self.is_reverse:
i = inputs.shape[1] - 1 - i i = inputs.shape[1] - 1 - i
input_ = inputs[ :, i:i+1, :] input_ = inputs[:, i:i + 1, :]
input_ = fluid.layers.reshape(input_, [-1, input_.shape[2]], inplace=False) input_ = fluid.layers.reshape(
input_, [-1, input_.shape[2]], inplace=False)
hidden, reset, gate = self.gru_unit(input_, hidden) hidden, reset, gate = self.gru_unit(input_, hidden)
hidden_ = fluid.layers.reshape(hidden, [-1, 1, hidden.shape[1]], inplace=False) hidden_ = fluid.layers.reshape(
hidden, [-1, 1, hidden.shape[1]], inplace=False)
res.append(hidden_) res.append(hidden_)
if self.is_reverse: if self.is_reverse:
res = res[::-1] res = res[::-1]
...@@ -786,10 +791,13 @@ class BasicLSTMUnit(RNNUnit): ...@@ -786,10 +791,13 @@ class BasicLSTMUnit(RNNUnit):
self._weight = self.create_parameter( self._weight = self.create_parameter(
attr=self._param_attr, attr=self._param_attr,
shape=[self._input_size + self._hidden_size, 4 * self._hidden_size], shape=[
self._input_size + self._hidden_size, 4 * self._hidden_size
],
dtype=self._dtype) dtype=self._dtype)
self._bias = self.create_parameter(attr=self._bias_attr, self._bias = self.create_parameter(
attr=self._bias_attr,
shape=[4 * self._hidden_size], shape=[4 * self._hidden_size],
dtype=self._dtype, dtype=self._dtype,
is_bias=True) is_bias=True)
...@@ -817,11 +825,7 @@ class BasicLSTMUnit(RNNUnit): ...@@ -817,11 +825,7 @@ class BasicLSTMUnit(RNNUnit):
class RNN(Layer): class RNN(Layer):
def __init__(self, def __init__(self, cell, is_reverse=False, time_major=False, **kwargs):
cell,
is_reverse=False,
time_major=False,
**kwargs):
super(RNN, self).__init__() super(RNN, self).__init__()
self.cell = cell self.cell = cell
if not hasattr(self.cell, "call"): if not hasattr(self.cell, "call"):
...@@ -831,12 +835,17 @@ class RNN(Layer): ...@@ -831,12 +835,17 @@ class RNN(Layer):
self.batch_index, self.time_step_index = (1, 0) if time_major else (0, self.batch_index, self.time_step_index = (1, 0) if time_major else (0,
1) 1)
def forward(self, inputs, initial_states=None, sequence_length=None, **kwargs): def forward(self,
inputs,
initial_states=None,
sequence_length=None,
**kwargs):
if fluid.in_dygraph_mode(): if fluid.in_dygraph_mode():
class OutputArray(object): class OutputArray(object):
def __init__(self, x): def __init__(self, x):
self.array = [x] self.array = [x]
def append(self, x): def append(self, x):
self.array.append(x) self.array.append(x)
...@@ -844,9 +853,8 @@ class RNN(Layer): ...@@ -844,9 +853,8 @@ class RNN(Layer):
# TODO: use where_op # TODO: use where_op
new_state = fluid.layers.elementwise_mul( new_state = fluid.layers.elementwise_mul(
new_state, step_mask, new_state, step_mask,
axis=0) - fluid.layers.elementwise_mul(state, axis=0) - fluid.layers.elementwise_mul(
(step_mask - 1), state, (step_mask - 1), axis=0)
axis=0)
return new_state return new_state
flat_inputs = flatten(inputs) flat_inputs = flatten(inputs)
...@@ -872,16 +880,20 @@ class RNN(Layer): ...@@ -872,16 +880,20 @@ class RNN(Layer):
if self.is_reverse: if self.is_reverse:
inputs = map_structure(lambda x: fluid.layers.reverse(x, axis=[0]), inputs) inputs = map_structure(lambda x: fluid.layers.reverse(x, axis=[0]), inputs)
mask = fluid.layers.reverse(mask, axis=[0]) if sequence_length is not None else None mask = fluid.layers.reverse(
mask, axis=[0]) if sequence_length is not None else None
states = initial_states states = initial_states
outputs = [] outputs = []
for i in range(time_steps): for i in range(time_steps):
step_inputs = map_structure(lambda x:x[i], inputs) step_inputs = map_structure(lambda x: x[i], inputs)
step_outputs, new_states = self.cell(step_inputs, states, **kwargs) step_outputs, new_states = self.cell(step_inputs, states,
**kwargs)
if sequence_length is not None: if sequence_length is not None:
new_states = map_structure( new_states = map_structure(
partial(_maybe_copy, step_mask=mask[i]), states, partial(
_maybe_copy, step_mask=mask[i]),
states,
new_states) new_states)
states = new_states states = new_states
if i == 0: if i == 0:
...@@ -922,10 +934,9 @@ class EncoderCell(RNNUnit): ...@@ -922,10 +934,9 @@ class EncoderCell(RNNUnit):
self.lstm_cells = list() self.lstm_cells = list()
for i in range(self.num_layers): for i in range(self.num_layers):
self.lstm_cells.append( self.lstm_cells.append(
self.add_sublayer( self.add_sublayer("layer_%d" % i,
"layer_%d" % i, BasicLSTMUnit(input_size if i == 0 else
BasicLSTMUnit(input_size if i == 0 else hidden_size, hidden_size, hidden_size)))
hidden_size)))
def forward(self, step_input, states): def forward(self, step_input, states):
new_states = [] new_states = []
...@@ -1040,4 +1051,3 @@ class BasicGRUUnit(Layer): ...@@ -1040,4 +1051,3 @@ class BasicGRUUnit(Layer):
new_hidden = u * pre_hidden + (1 - u) * c new_hidden = u * pre_hidden + (1 - u) * c
return new_hidden return new_hidden
dygraph/similarity_net/run_classifier.py
浏览文件 @ 1cb6a643
...@@ -85,7 +85,6 @@ def train(conf_dict, args): ...@@ -85,7 +85,6 @@ def train(conf_dict, args):
else: else:
return auc return auc
with fluid.dygraph.guard(place): with fluid.dygraph.guard(place):
# used for continuous evaluation # used for continuous evaluation
if args.enable_ce: if args.enable_ce:
...@@ -100,8 +99,7 @@ def train(conf_dict, args): ...@@ -100,8 +99,7 @@ def train(conf_dict, args):
conf_dict['seq_len'] = args.seq_len conf_dict['seq_len'] = args.seq_len
# Load network structure dynamically # Load network structure dynamically
net = utils.import_class("./nets", net = utils.import_class("./nets", conf_dict["net"]["module_name"],
conf_dict["net"]["module_name"],
conf_dict["net"]["class_name"])(conf_dict) conf_dict["net"]["class_name"])(conf_dict)
if args.init_checkpoint is not "": if args.init_checkpoint is not "":
model, _ = load_dygraph(args.init_checkpoint) model, _ = load_dygraph(args.init_checkpoint)
...@@ -113,9 +111,10 @@ def train(conf_dict, args): ...@@ -113,9 +111,10 @@ def train(conf_dict, args):
# Load Optimization method # Load Optimization method
learning_rate = conf_dict["optimizer"]["learning_rate"] learning_rate = conf_dict["optimizer"]["learning_rate"]
optimizer_name = conf_dict["optimizer"]["class_name"] optimizer_name = conf_dict["optimizer"]["class_name"]
if optimizer_name=='SGDOptimizer': if optimizer_name == 'SGDOptimizer':
optimizer = fluid.optimizer.SGDOptimizer(learning_rate,parameter_list=net.parameters()) optimizer = fluid.optimizer.SGDOptimizer(
elif optimizer_name=='AdamOptimizer': learning_rate, parameter_list=net.parameters())
elif optimizer_name == 'AdamOptimizer':
beta1 = conf_dict["optimizer"]["beta1"] beta1 = conf_dict["optimizer"]["beta1"]
beta2 = conf_dict["optimizer"]["beta2"] beta2 = conf_dict["optimizer"]["beta2"]
epsilon = conf_dict["optimizer"]["epsilon"] epsilon = conf_dict["optimizer"]["epsilon"]
...@@ -136,22 +135,32 @@ def train(conf_dict, args): ...@@ -136,22 +135,32 @@ def train(conf_dict, args):
losses = [] losses = []
start_time = time.time() start_time = time.time()
train_pyreader = fluid.io.PyReader(capacity=16, return_list=True, use_double_buffer=True) train_loader = fluid.io.DataLoader.from_generator(
get_train_examples = simnet_process.get_reader("train",epoch=args.epoch) capacity=16,
train_pyreader.decorate_sample_list_generator( return_list=True,
paddle.batch(get_train_examples, batch_size=args.batch_size), iterable=True,
place) use_double_buffer=True)
get_train_examples = simnet_process.get_reader(
"train", epoch=args.epoch)
train_loader.set_sample_list_generator(
paddle.batch(
get_train_examples, batch_size=args.batch_size), place)
if args.do_valid: if args.do_valid:
valid_pyreader = fluid.io.PyReader(capacity=16, return_list=True, use_double_buffer=True) valid_loader = fluid.io.DataLoader.from_generator(
capacity=16,
return_list=True,
iterable=True,
use_double_buffer=True)
get_valid_examples = simnet_process.get_reader("valid") get_valid_examples = simnet_process.get_reader("valid")
valid_pyreader.decorate_sample_list_generator( valid_loader.set_sample_list_generator(
paddle.batch(get_valid_examples, batch_size=args.batch_size), paddle.batch(
get_valid_examples, batch_size=args.batch_size),
place) place)
pred_list = [] pred_list = []
if args.task_mode == "pairwise": if args.task_mode == "pairwise":
for left, pos_right, neg_right in train_pyreader(): for left, pos_right, neg_right in train_loader():
left = fluid.layers.reshape(left, shape=[-1, 1]) left = fluid.layers.reshape(left, shape=[-1, 1])
pos_right = fluid.layers.reshape(pos_right, shape=[-1, 1]) pos_right = fluid.layers.reshape(pos_right, shape=[-1, 1])
...@@ -168,23 +177,27 @@ def train(conf_dict, args): ...@@ -168,23 +177,27 @@ def train(conf_dict, args):
net.clear_gradients() net.clear_gradients()
if args.do_valid and global_step % args.validation_steps == 0: if args.do_valid and global_step % args.validation_steps == 0:
for left, pos_right in valid_pyreader(): for left, pos_right in valid_loader():
left = fluid.layers.reshape(left, shape=[-1, 1]) left = fluid.layers.reshape(left, shape=[-1, 1])
pos_right = fluid.layers.reshape(pos_right, shape=[-1, 1]) pos_right = fluid.layers.reshape(
pos_right, shape=[-1, 1])
net.eval() net.eval()
left_feat, pos_score = net(left, pos_right) left_feat, pos_score = net(left, pos_right)
pred = pos_score pred = pos_score
pred_list += list(pred.numpy()) pred_list += list(pred.numpy())
valid_result = valid_and_test(pred_list, simnet_process, "valid") valid_result = valid_and_test(pred_list, simnet_process,
"valid")
if args.compute_accuracy: if args.compute_accuracy:
valid_auc, valid_acc = valid_result valid_auc, valid_acc = valid_result
logging.info( logging.info(
"global_steps: %d, valid_auc: %f, valid_acc: %f, valid_loss: %f" % "global_steps: %d, valid_auc: %f, valid_acc: %f, valid_loss: %f"
(global_step, valid_auc, valid_acc, np.mean(losses))) % (global_step, valid_auc, valid_acc,
np.mean(losses)))
else: else:
valid_auc = valid_result valid_auc = valid_result
logging.info("global_steps: %d, valid_auc: %f, valid_loss: %f" % logging.info(
"global_steps: %d, valid_auc: %f, valid_loss: %f" %
(global_step, valid_auc, np.mean(losses))) (global_step, valid_auc, np.mean(losses)))
if global_step % args.save_steps == 0: if global_step % args.save_steps == 0:
...@@ -198,7 +211,7 @@ def train(conf_dict, args): ...@@ -198,7 +211,7 @@ def train(conf_dict, args):
logging.info("saving infer model in %s" % model_path) logging.info("saving infer model in %s" % model_path)
else: else:
for left, right, label in train_pyreader(): for left, right, label in train_loader():
left = fluid.layers.reshape(left, shape=[-1, 1]) left = fluid.layers.reshape(left, shape=[-1, 1])
right = fluid.layers.reshape(right, shape=[-1, 1]) right = fluid.layers.reshape(right, shape=[-1, 1])
label = fluid.layers.reshape(label, shape=[-1, 1]) label = fluid.layers.reshape(label, shape=[-1, 1])
...@@ -212,21 +225,24 @@ def train(conf_dict, args): ...@@ -212,21 +225,24 @@ def train(conf_dict, args):
net.clear_gradients() net.clear_gradients()
if args.do_valid and global_step % args.validation_steps == 0: if args.do_valid and global_step % args.validation_steps == 0:
for left, right in valid_pyreader(): for left, right in valid_loader():
left = fluid.layers.reshape(left, shape=[-1, 1]) left = fluid.layers.reshape(left, shape=[-1, 1])
right = fluid.layers.reshape(right, shape=[-1, 1]) right = fluid.layers.reshape(right, shape=[-1, 1])
net.eval() net.eval()
left_feat, pred = net(left, right) left_feat, pred = net(left, right)
pred_list += list(pred.numpy()) pred_list += list(pred.numpy())
valid_result = valid_and_test(pred_list, simnet_process, "valid") valid_result = valid_and_test(pred_list, simnet_process,
"valid")
if args.compute_accuracy: if args.compute_accuracy:
valid_auc, valid_acc = valid_result valid_auc, valid_acc = valid_result
logging.info( logging.info(
"global_steps: %d, valid_auc: %f, valid_acc: %f, valid_loss: %f" % "global_steps: %d, valid_auc: %f, valid_acc: %f, valid_loss: %f"
(global_step, valid_auc, valid_acc, np.mean(losses))) % (global_step, valid_auc, valid_acc,
np.mean(losses)))
else: else:
valid_auc = valid_result valid_auc = valid_result
logging.info("global_steps: %d, valid_auc: %f, valid_loss: %f" % logging.info(
"global_steps: %d, valid_auc: %f, valid_loss: %f" %
(global_step, valid_auc, np.mean(losses))) (global_step, valid_auc, np.mean(losses)))
if global_step % args.save_steps == 0: if global_step % args.save_steps == 0:
...@@ -244,8 +260,7 @@ def train(conf_dict, args): ...@@ -244,8 +260,7 @@ def train(conf_dict, args):
ce_info.append([np.mean(losses), end_time - start_time]) ce_info.append([np.mean(losses), end_time - start_time])
# final save # final save
logging.info("the final step is %s" % global_step) logging.info("the final step is %s" % global_step)
model_save_dir = os.path.join(args.output_dir, model_save_dir = os.path.join(args.output_dir, conf_dict["model_path"])
conf_dict["model_path"])
model_path = os.path.join(model_save_dir, str(global_step)) model_path = os.path.join(model_save_dir, str(global_step))
if not os.path.exists(model_save_dir): if not os.path.exists(model_save_dir):
...@@ -269,13 +284,18 @@ def train(conf_dict, args): ...@@ -269,13 +284,18 @@ def train(conf_dict, args):
if args.do_test: if args.do_test:
# Get Feeder and Reader # Get Feeder and Reader
test_pyreader = fluid.io.PyReader(capacity=16, return_list=True, use_double_buffer=True) test_loader = fluid.io.DataLoader.from_generator(
capacity=16,
return_list=True,
iterable=True,
use_double_buffer=True)
get_test_examples = simnet_process.get_reader("test") get_test_examples = simnet_process.get_reader("test")
test_pyreader.decorate_sample_list_generator( test_loader.set_sample_list_generator(
paddle.batch(get_test_examples, batch_size=args.batch_size), paddle.batch(
get_test_examples, batch_size=args.batch_size),
place) place)
pred_list = [] pred_list = []
for left, pos_right in test_pyreader(): for left, pos_right in test_loader():
left = fluid.layers.reshape(left, shape=[-1, 1]) left = fluid.layers.reshape(left, shape=[-1, 1])
pos_right = fluid.layers.reshape(pos_right, shape=[-1, 1]) pos_right = fluid.layers.reshape(pos_right, shape=[-1, 1])
net.eval() net.eval()
...@@ -307,47 +327,53 @@ def test(conf_dict, args): ...@@ -307,47 +327,53 @@ def test(conf_dict, args):
vocab = utils.load_vocab(args.vocab_path) vocab = utils.load_vocab(args.vocab_path)
simnet_process = reader.SimNetProcessor(args, vocab) simnet_process = reader.SimNetProcessor(args, vocab)
test_pyreader = fluid.io.PyReader(capacity=16, return_list=True, use_double_buffer=True) test_loader = fluid.io.DataLoader.from_generator(
capacity=16,
return_list=True,
iterable=True,
use_double_buffer=True)
get_test_examples = simnet_process.get_reader("test") get_test_examples = simnet_process.get_reader("test")
test_pyreader.decorate_sample_list_generator( test_loader.set_sample_list_generator(
paddle.batch(get_test_examples, batch_size=args.batch_size), paddle.batch(
place) get_test_examples, batch_size=args.batch_size), place)
conf_dict['dict_size'] = len(vocab) conf_dict['dict_size'] = len(vocab)
conf_dict['seq_len'] = args.seq_len conf_dict['seq_len'] = args.seq_len
net = utils.import_class("./nets", net = utils.import_class("./nets", conf_dict["net"]["module_name"],
conf_dict["net"]["module_name"],
conf_dict["net"]["class_name"])(conf_dict) conf_dict["net"]["class_name"])(conf_dict)
model, _ = load_dygraph(args.init_checkpoint) model, _ = load_dygraph(args.init_checkpoint)
net.set_dict(model) net.set_dict(model)
metric = fluid.metrics.Auc(name="auc") metric = fluid.metrics.Auc(name="auc")
pred_list = [] pred_list = []
with io.open("predictions.txt", "w", encoding="utf8") as predictions_file: with io.open(
"predictions.txt", "w", encoding="utf8") as predictions_file:
if args.task_mode == "pairwise": if args.task_mode == "pairwise":
for left, pos_right in test_pyreader(): for left, pos_right in test_loader():
left = fluid.layers.reshape(left, shape=[-1, 1]) left = fluid.layers.reshape(left, shape=[-1, 1])
pos_right = fluid.layers.reshape(pos_right, shape=[-1, 1]) pos_right = fluid.layers.reshape(pos_right, shape=[-1, 1])
left_feat, pos_score = net(left, pos_right) left_feat, pos_score = net(left, pos_right)
pred = pos_score pred = pos_score
pred_list += list(map(lambda item: float(item[0]), pred.numpy())) pred_list += list(
map(lambda item: float(item[0]), pred.numpy()))
predictions_file.write(u"\n".join( predictions_file.write(u"\n".join(
map(lambda item: str((item[0] + 1) / 2), pred.numpy())) + "\n") map(lambda item: str((item[0] + 1) / 2), pred.numpy()))
+ "\n")
else: else:
for left, right in test_pyreader(): for left, right in test_loader():
left = fluid.layers.reshape(left, shape=[-1, 1]) left = fluid.layers.reshape(left, shape=[-1, 1])
right = fluid.layers.reshape(right, shape=[-1, 1]) right = fluid.layers.reshape(right, shape=[-1, 1])
left_feat, pred = net(left, right) left_feat, pred = net(left, right)
pred_list += list(map(lambda item: float(item[0]), pred.numpy())) pred_list += list(
map(lambda item: float(item[0]), pred.numpy()))
predictions_file.write(u"\n".join( predictions_file.write(u"\n".join(
map(lambda item: str(np.argmax(item)), pred.numpy())) + "\n") map(lambda item: str(np.argmax(item)), pred.numpy())) +
"\n")
if args.task_mode == "pairwise": if args.task_mode == "pairwise":
pred_list = np.array(pred_list).reshape((-1, 1)) pred_list = np.array(pred_list).reshape((-1, 1))
...@@ -383,28 +409,30 @@ def infer(conf_dict, args): ...@@ -383,28 +409,30 @@ def infer(conf_dict, args):
else: else:
place = fluid.CPUPlace() place = fluid.CPUPlace()
with fluid.dygraph.guard(place): with fluid.dygraph.guard(place):
vocab = utils.load_vocab(args.vocab_path) vocab = utils.load_vocab(args.vocab_path)
simnet_process = reader.SimNetProcessor(args, vocab) simnet_process = reader.SimNetProcessor(args, vocab)
get_infer_examples = simnet_process.get_infer_reader get_infer_examples = simnet_process.get_infer_reader
infer_pyreader = fluid.io.PyReader(capacity=16, return_list=True, use_double_buffer=True) infer_loader = fluid.io.DataLoader.from_generator(
infer_pyreader.decorate_sample_list_generator( capacity=16,
paddle.batch(get_infer_examples, batch_size=args.batch_size), return_list=True,
place) iterable=True,
use_double_buffer=True)
infer_loader.set_sample_list_generator(
paddle.batch(
get_infer_examples, batch_size=args.batch_size), place)
conf_dict['dict_size'] = len(vocab) conf_dict['dict_size'] = len(vocab)
conf_dict['seq_len'] = args.seq_len conf_dict['seq_len'] = args.seq_len
net = utils.import_class("./nets", net = utils.import_class("./nets", conf_dict["net"]["module_name"],
conf_dict["net"]["module_name"],
conf_dict["net"]["class_name"])(conf_dict) conf_dict["net"]["class_name"])(conf_dict)
model, _ = load_dygraph(args.init_checkpoint) model, _ = load_dygraph(args.init_checkpoint)
net.set_dict(model) net.set_dict(model)
pred_list = [] pred_list = []
if args.task_mode == "pairwise": if args.task_mode == "pairwise":
for left, pos_right in infer_pyreader(): for left, pos_right in infer_loader():
left = fluid.layers.reshape(left, shape=[-1, 1]) left = fluid.layers.reshape(left, shape=[-1, 1])
pos_right = fluid.layers.reshape(pos_right, shape=[-1, 1]) pos_right = fluid.layers.reshape(pos_right, shape=[-1, 1])
...@@ -414,14 +442,15 @@ def infer(conf_dict, args): ...@@ -414,14 +442,15 @@ def infer(conf_dict, args):
map(lambda item: str((item[0] + 1) / 2), pred.numpy())) map(lambda item: str((item[0] + 1) / 2), pred.numpy()))
else: else:
for left, right in infer_pyreader(): for left, right in infer_loader():
left = fluid.layers.reshape(left, shape=[-1, 1]) left = fluid.layers.reshape(left, shape=[-1, 1])
pos_right = fluid.layers.reshape(right, shape=[-1, 1]) pos_right = fluid.layers.reshape(right, shape=[-1, 1])
left_feat, pred = net(left, right) left_feat, pred = net(left, right)
pred_list += map(lambda item: str(np.argmax(item)), pred.numpy()) pred_list += map(lambda item: str(np.argmax(item)),
pred.numpy())
with io.open(
with io.open(args.infer_result_path, "w", encoding="utf8") as infer_file: args.infer_result_path, "w", encoding="utf8") as infer_file:
for _data, _pred in zip(simnet_process.get_infer_data(), pred_list): for _data, _pred in zip(simnet_process.get_infer_data(), pred_list):
infer_file.write(_data + "\t" + _pred + "\n") infer_file.write(_data + "\t" + _pred + "\n")
logging.info("infer result saved in %s" % logging.info("infer result saved in %s" %
...@@ -435,6 +464,7 @@ def get_cards(): ...@@ -435,6 +464,7 @@ def get_cards():
num = len(cards.split(",")) num = len(cards.split(","))
return num return num
if __name__ == "__main__": if __name__ == "__main__":
args = ArgConfig() args = ArgConfig()
......
dygraph/transformer/README.md
浏览文件 @ 1cb6a643
...@@ -28,7 +28,7 @@ ...@@ -28,7 +28,7 @@
1. paddle安装 1. paddle安装
本项目依赖于 PaddlePaddle 1.7及以上版本或适当的develop版本，请参考 [安装指南](http://www.paddlepaddle.org/#quick-start) 进行安装 本项目依赖于 PaddlePaddle 1.8及以上版本或适当的develop版本，请参考 [安装指南](https://www.paddlepaddle.org.cn/install/quick) 进行安装
2. 下载代码 2. 下载代码
...@@ -40,7 +40,7 @@ ...@@ -40,7 +40,7 @@
3. 环境依赖 3. 环境依赖
请参考PaddlePaddle[安装说明](https://www.paddlepaddle.org.cn/documentation/docs/zh/1.6/beginners_guide/install/index_cn.html)部分的内容 请参考PaddlePaddle[安装说明](https://www.paddlepaddle.org.cn/documentation/docs/zh/install/index_cn.html)部分的内容
### 数据准备 ### 数据准备
......
dygraph/transformer/config.py
浏览文件 @ 1cb6a643
...@@ -42,12 +42,11 @@ class InferTaskConfig(object): ...@@ -42,12 +42,11 @@ class InferTaskConfig(object):
batch_size = 4 batch_size = 4
# the parameters for beam search. # the parameters for beam search.
beam_size = 4 beam_size = 4
alpha=0.6 alpha = 0.6
# max decoded length, should be less than ModelHyperParams.max_length # max decoded length, should be less than ModelHyperParams.max_length
max_out_len = 30 max_out_len = 30
class ModelHyperParams(object): class ModelHyperParams(object):
""" """
ModelHyperParams ModelHyperParams
...@@ -156,38 +155,32 @@ input_descs = { ...@@ -156,38 +155,32 @@ input_descs = {
# Names of word embedding table which might be reused for weight sharing. # Names of word embedding table which might be reused for weight sharing.
word_emb_param_names = ( word_emb_param_names = (
"src_word_emb_table", "src_word_emb_table",
"trg_word_emb_table", "trg_word_emb_table", )
)
# Names of position encoding table which will be initialized externally. # Names of position encoding table which will be initialized externally.
pos_enc_param_names = ( pos_enc_param_names = (
"src_pos_enc_table", "src_pos_enc_table",
"trg_pos_enc_table", "trg_pos_enc_table", )
)
# separated inputs for different usages. # separated inputs for different usages.
encoder_data_input_fields = ( encoder_data_input_fields = (
"src_word", "src_word",
"src_pos", "src_pos",
"src_slf_attn_bias", "src_slf_attn_bias", )
)
decoder_data_input_fields = ( decoder_data_input_fields = (
"trg_word", "trg_word",
"trg_pos", "trg_pos",
"trg_slf_attn_bias", "trg_slf_attn_bias",
"trg_src_attn_bias", "trg_src_attn_bias",
"enc_output", "enc_output", )
)
label_data_input_fields = ( label_data_input_fields = (
"lbl_word", "lbl_word",
"lbl_weight", "lbl_weight", )
)
# In fast decoder, trg_pos (only containing the current time step) is generated # In fast decoder, trg_pos (only containing the current time step) is generated
# by ops and trg_slf_attn_bias is not needed. # by ops and trg_slf_attn_bias is not needed.
fast_decoder_data_input_fields = ( fast_decoder_data_input_fields = (
"trg_word", "trg_word",
# "init_score", # "init_score",
# "init_idx", # "init_idx",
"trg_src_attn_bias", "trg_src_attn_bias", )
)
def merge_cfg_from_list(cfg_list, g_cfgs): def merge_cfg_from_list(cfg_list, g_cfgs):
......
dygraph/transformer/model.py
浏览文件 @ 1cb6a643
此差异已折叠。
dygraph/transformer/reader.py
浏览文件 @ 1cb6a643
...@@ -306,6 +306,7 @@ class DataProcessor(object): ...@@ -306,6 +306,7 @@ class DataProcessor(object):
:param seed: The seed for random. :param seed: The seed for random.
:type seed: int :type seed: int
""" """
def __init__(self, def __init__(self,
src_vocab_fpath, src_vocab_fpath,
trg_vocab_fpath, trg_vocab_fpath,
...@@ -360,7 +361,8 @@ class DataProcessor(object): ...@@ -360,7 +361,8 @@ class DataProcessor(object):
def load_src_trg_ids(self, fpattern, tar_fname): def load_src_trg_ids(self, fpattern, tar_fname):
converters = [ converters = [
Converter(vocab=self._src_vocab, Converter(
vocab=self._src_vocab,
beg=self._bos_idx, beg=self._bos_idx,
end=self._eos_idx, end=self._eos_idx,
unk=self._unk_idx, unk=self._unk_idx,
...@@ -369,7 +371,8 @@ class DataProcessor(object): ...@@ -369,7 +371,8 @@ class DataProcessor(object):
] ]
if not self._only_src: if not self._only_src:
converters.append( converters.append(
Converter(vocab=self._trg_vocab, Converter(
vocab=self._trg_vocab,
beg=self._bos_idx, beg=self._bos_idx,
end=self._eos_idx, end=self._eos_idx,
unk=self._unk_idx, unk=self._unk_idx,
...@@ -402,9 +405,8 @@ class DataProcessor(object): ...@@ -402,9 +405,8 @@ class DataProcessor(object):
f = tarfile.open(fpaths[0], "rb") f = tarfile.open(fpaths[0], "rb")
for line in f.extractfile(tar_fname): for line in f.extractfile(tar_fname):
fields = line.strip(b"\n").split(self._field_delimiter) fields = line.strip(b"\n").split(self._field_delimiter)
if (not self._only_src if (not self._only_src and len(fields) == 2) or (
and len(fields) == 2) or (self._only_src self._only_src and len(fields) == 1):
and len(fields) == 1):
yield fields yield fields
else: else:
for fpath in fpaths: for fpath in fpaths:
...@@ -414,9 +416,8 @@ class DataProcessor(object): ...@@ -414,9 +416,8 @@ class DataProcessor(object):
with open(fpath, "rb") as f: with open(fpath, "rb") as f:
for line in f: for line in f:
fields = line.strip(b"\n").split(self._field_delimiter) fields = line.strip(b"\n").split(self._field_delimiter)
if (not self._only_src if (not self._only_src and len(fields) == 2) or (
and len(fields) == 2) or (self._only_src self._only_src and len(fields) == 1):
and len(fields) == 1):
yield fields yield fields
@staticmethod @staticmethod
...@@ -512,8 +513,8 @@ class DataProcessor(object): ...@@ -512,8 +513,8 @@ class DataProcessor(object):
for item in data_reader(): for item in data_reader():
inst_num_per_part = len(item) // count inst_num_per_part = len(item) // count
for i in range(count): for i in range(count):
yield item[inst_num_per_part * i:inst_num_per_part * yield item[inst_num_per_part * i:inst_num_per_part * (i + 1
(i + 1)] )]
return __impl__ return __impl__
......
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