Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
PaddlePaddle
models
提交
4c681dfb
M
models
项目概览
PaddlePaddle
/
models
大约 2 年 前同步成功
通知
232
Star
6828
Fork
2962
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
602
列表
看板
标记
里程碑
合并请求
255
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
M
models
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
602
Issue
602
列表
看板
标记
里程碑
合并请求
255
合并请求
255
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
提交
4c681dfb
编写于
11月 11, 2018
作者:
Y
Yibing Liu
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Fix test & feed type and disable grad clip temporarily
上级
8c5f9d7e
变更
3
隐藏空白更改
内联
并排
Showing
3 changed file
with
13 addition
and
11 deletion
+13
-11
fluid/PaddleNLP/deep_attention_matching_net/model.py
fluid/PaddleNLP/deep_attention_matching_net/model.py
+6
-6
fluid/PaddleNLP/deep_attention_matching_net/test_and_evaluate.py
...addleNLP/deep_attention_matching_net/test_and_evaluate.py
+4
-2
fluid/PaddleNLP/deep_attention_matching_net/train_and_evaluate.py
...ddleNLP/deep_attention_matching_net/train_and_evaluate.py
+3
-3
未找到文件。
fluid/PaddleNLP/deep_attention_matching_net/model.py
浏览文件 @
4c681dfb
...
@@ -25,7 +25,7 @@ class Net(object):
...
@@ -25,7 +25,7 @@ class Net(object):
# turns ids
# turns ids
shapes
=
[[
-
1
,
self
.
_max_turn_len
,
1
]
shapes
=
[[
-
1
,
self
.
_max_turn_len
,
1
]
for
i
in
six
.
moves
.
xrange
(
self
.
_max_turn_num
)]
for
i
in
six
.
moves
.
xrange
(
self
.
_max_turn_num
)]
dtypes
=
[
"int
32
"
for
i
in
six
.
moves
.
xrange
(
self
.
_max_turn_num
)]
dtypes
=
[
"int
64
"
for
i
in
six
.
moves
.
xrange
(
self
.
_max_turn_num
)]
# turns mask
# turns mask
shapes
+=
[[
-
1
,
self
.
_max_turn_len
,
1
]
shapes
+=
[[
-
1
,
self
.
_max_turn_len
,
1
]
for
i
in
six
.
moves
.
xrange
(
self
.
_max_turn_num
)]
for
i
in
six
.
moves
.
xrange
(
self
.
_max_turn_num
)]
...
@@ -34,7 +34,7 @@ class Net(object):
...
@@ -34,7 +34,7 @@ class Net(object):
# response ids, response mask, label
# response ids, response mask, label
shapes
+=
[[
-
1
,
self
.
_max_turn_len
,
1
],
[
-
1
,
self
.
_max_turn_len
,
1
],
shapes
+=
[[
-
1
,
self
.
_max_turn_len
,
1
],
[
-
1
,
self
.
_max_turn_len
,
1
],
[
-
1
,
1
]]
[
-
1
,
1
]]
dtypes
+=
[
"int
32
"
,
"float32"
,
"float32"
]
dtypes
+=
[
"int
64
"
,
"float32"
,
"float32"
]
py_reader
=
fluid
.
layers
.
py_reader
(
py_reader
=
fluid
.
layers
.
py_reader
(
capacity
=
capacity
,
capacity
=
capacity
,
...
@@ -60,7 +60,7 @@ class Net(object):
...
@@ -60,7 +60,7 @@ class Net(object):
for
i
in
six
.
moves
.
xrange
(
self
.
_max_turn_num
):
for
i
in
six
.
moves
.
xrange
(
self
.
_max_turn_num
):
name
=
"turn_%d"
%
i
name
=
"turn_%d"
%
i
turn
=
fluid
.
layers
.
data
(
turn
=
fluid
.
layers
.
data
(
name
=
name
,
shape
=
[
self
.
_max_turn_len
,
1
],
dtype
=
"int
32
"
)
name
=
name
,
shape
=
[
self
.
_max_turn_len
,
1
],
dtype
=
"int
64
"
)
self
.
turns_data
.
append
(
turn
)
self
.
turns_data
.
append
(
turn
)
self
.
_feed_names
.
append
(
name
)
self
.
_feed_names
.
append
(
name
)
...
@@ -73,7 +73,7 @@ class Net(object):
...
@@ -73,7 +73,7 @@ class Net(object):
self
.
_feed_names
.
append
(
name
)
self
.
_feed_names
.
append
(
name
)
self
.
response
=
fluid
.
layers
.
data
(
self
.
response
=
fluid
.
layers
.
data
(
name
=
"response"
,
shape
=
[
self
.
_max_turn_len
,
1
],
dtype
=
"int
32
"
)
name
=
"response"
,
shape
=
[
self
.
_max_turn_len
,
1
],
dtype
=
"int
64
"
)
self
.
response_mask
=
fluid
.
layers
.
data
(
self
.
response_mask
=
fluid
.
layers
.
data
(
name
=
"response_mask"
,
name
=
"response_mask"
,
shape
=
[
self
.
_max_turn_len
,
1
],
shape
=
[
self
.
_max_turn_len
,
1
],
...
@@ -141,7 +141,7 @@ class Net(object):
...
@@ -141,7 +141,7 @@ class Net(object):
mask_cache
=
mask_cache
)
mask_cache
=
mask_cache
)
Hu_stack
.
append
(
Hu
)
Hu_stack
.
append
(
Hu
)
# cross attention
# cross attention
r_a_t_stack
=
[]
r_a_t_stack
=
[]
t_a_r_stack
=
[]
t_a_r_stack
=
[]
for
index
in
six
.
moves
.
xrange
(
self
.
_stack_num
+
1
):
for
index
in
six
.
moves
.
xrange
(
self
.
_stack_num
+
1
):
...
@@ -183,7 +183,7 @@ class Net(object):
...
@@ -183,7 +183,7 @@ class Net(object):
t_a_r
=
fluid
.
layers
.
concat
(
input
=
t_a_r_stack
,
axis
=
1
)
t_a_r
=
fluid
.
layers
.
concat
(
input
=
t_a_r_stack
,
axis
=
1
)
r_a_t
=
fluid
.
layers
.
concat
(
input
=
r_a_t_stack
,
axis
=
1
)
r_a_t
=
fluid
.
layers
.
concat
(
input
=
r_a_t_stack
,
axis
=
1
)
# sim shape: [batch_size, 2*(stack_num+1), max_turn_len, max_turn_len]
# sim shape: [batch_size, 2*(stack_num+1), max_turn_len, max_turn_len]
sim
=
fluid
.
layers
.
matmul
(
sim
=
fluid
.
layers
.
matmul
(
x
=
t_a_r
,
y
=
r_a_t
,
transpose_y
=
True
,
alpha
=
1
/
np
.
sqrt
(
200.0
))
x
=
t_a_r
,
y
=
r_a_t
,
transpose_y
=
True
,
alpha
=
1
/
np
.
sqrt
(
200.0
))
sim_turns
.
append
(
sim
)
sim_turns
.
append
(
sim
)
...
...
fluid/PaddleNLP/deep_attention_matching_net/test_and_evaluate.py
浏览文件 @
4c681dfb
...
@@ -126,6 +126,7 @@ def test(args):
...
@@ -126,6 +126,7 @@ def test(args):
dam
=
Net
(
args
.
max_turn_num
,
args
.
max_turn_len
,
args
.
vocab_size
,
dam
=
Net
(
args
.
max_turn_num
,
args
.
max_turn_len
,
args
.
vocab_size
,
args
.
emb_size
,
args
.
stack_num
,
args
.
channel1_num
,
args
.
emb_size
,
args
.
stack_num
,
args
.
channel1_num
,
args
.
channel2_num
)
args
.
channel2_num
)
dam
.
create_data_layers
()
loss
,
logits
=
dam
.
create_network
()
loss
,
logits
=
dam
.
create_network
()
loss
.
persistable
=
True
loss
.
persistable
=
True
...
@@ -144,7 +145,7 @@ def test(args):
...
@@ -144,7 +145,7 @@ def test(args):
staircase
=
True
))
staircase
=
True
))
optimizer
.
minimize
(
loss
)
optimizer
.
minimize
(
loss
)
# The fethced loss is wrong when mem opt is enabled
# The fethced loss is wrong when mem opt is enabled
fluid
.
memory_optimize
(
fluid
.
default_main_program
())
fluid
.
memory_optimize
(
fluid
.
default_main_program
())
if
args
.
use_cuda
:
if
args
.
use_cuda
:
...
@@ -191,7 +192,8 @@ def test(args):
...
@@ -191,7 +192,8 @@ def test(args):
feed_list
=
[]
feed_list
=
[]
for
dev
in
six
.
moves
.
xrange
(
dev_count
):
for
dev
in
six
.
moves
.
xrange
(
dev_count
):
index
=
it
*
dev_count
+
dev
index
=
it
*
dev_count
+
dev
feed_dict
=
reader
.
make_one_batch_input
(
test_batches
,
index
)
batch_data
=
reader
.
make_one_batch_input
(
test_batches
,
index
)
feed_dict
=
dict
(
zip
(
dam
.
get_feed_names
(),
batch_data
))
feed_list
.
append
(
feed_dict
)
feed_list
.
append
(
feed_dict
)
predicts
=
test_exe
.
run
(
feed
=
feed_list
,
fetch_list
=
[
logits
.
name
])
predicts
=
test_exe
.
run
(
feed
=
feed_list
,
fetch_list
=
[
logits
.
name
])
...
...
fluid/PaddleNLP/deep_attention_matching_net/train_and_evaluate.py
浏览文件 @
4c681dfb
...
@@ -203,8 +203,8 @@ def train(args):
...
@@ -203,8 +203,8 @@ def train(args):
loss
.
persistable
=
True
loss
.
persistable
=
True
logits
.
persistable
=
True
logits
.
persistable
=
True
# gradient clipping
# gradient clipping
fluid
.
clip
.
set_gradient_clip
(
clip
=
fluid
.
clip
.
GradientClipByValue
(
#
fluid.clip.set_gradient_clip(clip=fluid.clip.GradientClipByValue(
max
=
1.0
,
min
=-
1.0
))
#
max=1.0, min=-1.0))
optimizer
=
fluid
.
optimizer
.
Adam
(
optimizer
=
fluid
.
optimizer
.
Adam
(
learning_rate
=
fluid
.
layers
.
exponential_decay
(
learning_rate
=
fluid
.
layers
.
exponential_decay
(
...
@@ -324,7 +324,7 @@ def train(args):
...
@@ -324,7 +324,7 @@ def train(args):
evaluate
(
score_path
,
result_file_path
)
evaluate
(
score_path
,
result_file_path
)
return
step
return
step
# train on one epoch with pyreader
# train on one epoch with pyreader
def
train_with_pyreader
(
step
):
def
train_with_pyreader
(
step
):
def
data_provider
():
def
data_provider
():
for
index
in
six
.
moves
.
xrange
(
batch_num
):
for
index
in
six
.
moves
.
xrange
(
batch_num
):
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录