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
兽拳
Paddle
提交
f364b722
P
Paddle
项目概览
兽拳
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
1
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1
列表
看板
标记
里程碑
合并请求
0
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
Paddle
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
1
Issue
1
列表
看板
标记
里程碑
合并请求
0
合并请求
0
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
体验新版 GitCode,发现更多精彩内容 >>
提交
f364b722
编写于
1月 25, 2019
作者:
J
JiabinYang
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
test=develop, add ptb_rnn test in imperative
上级
a59b7ac7
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
120 addition
and
50 deletion
+120
-50
python/paddle/fluid/tests/unittests/test_imperative_ptb_rnn.py
...n/paddle/fluid/tests/unittests/test_imperative_ptb_rnn.py
+120
-49
python/paddle/fluid/tests/unittests/test_imperative_split.py
python/paddle/fluid/tests/unittests/test_imperative_split.py
+0
-1
未找到文件。
python/paddle/fluid/tests/unittests/test_imperative_ptb_rnn.py
浏览文件 @
f364b722
...
@@ -20,7 +20,9 @@ from paddle.fluid.imperative.nn import EMBEDDING
...
@@ -20,7 +20,9 @@ from paddle.fluid.imperative.nn import EMBEDDING
import
paddle.fluid.framework
as
framework
import
paddle.fluid.framework
as
framework
from
paddle.fluid.optimizer
import
SGDOptimizer
from
paddle.fluid.optimizer
import
SGDOptimizer
from
paddle.fluid.imperative.base
import
to_variable
from
paddle.fluid.imperative.base
import
to_variable
from
test_imperative_base
import
new_program_scope
import
numpy
as
np
import
numpy
as
np
import
six
from
paddle.fluid.backward
import
append_backward
from
paddle.fluid.backward
import
append_backward
...
@@ -36,8 +38,8 @@ class SimpleLSTMRNN(fluid.imperative.Layer):
...
@@ -36,8 +38,8 @@ class SimpleLSTMRNN(fluid.imperative.Layer):
self
.
_num_layers
=
num_layers
self
.
_num_layers
=
num_layers
self
.
_init_scale
=
init_scale
self
.
_init_scale
=
init_scale
self
.
_dropout
=
dropout
self
.
_dropout
=
dropout
self
.
input
=
None
self
.
_
input
=
None
self
.
num_steps
=
num_steps
self
.
_
num_steps
=
num_steps
def
_build_once
(
self
,
input_embedding
,
init_hidden
=
None
,
init_cell
=
None
):
def
_build_once
(
self
,
input_embedding
,
init_hidden
=
None
,
init_cell
=
None
):
self
.
weight_1_arr
=
[]
self
.
weight_1_arr
=
[]
...
@@ -75,58 +77,49 @@ class SimpleLSTMRNN(fluid.imperative.Layer):
...
@@ -75,58 +77,49 @@ class SimpleLSTMRNN(fluid.imperative.Layer):
def
forward
(
self
,
input_embedding
,
init_hidden
=
None
,
init_cell
=
None
):
def
forward
(
self
,
input_embedding
,
init_hidden
=
None
,
init_cell
=
None
):
res
=
[]
res
=
[]
for
index
in
range
(
self
.
num_steps
):
for
index
in
range
(
self
.
_
num_steps
):
self
.
input
=
fluid
.
layers
.
slice
(
self
.
_
input
=
fluid
.
layers
.
slice
(
input_embedding
,
axes
=
[
1
],
starts
=
[
index
],
ends
=
[
index
+
1
])
input_embedding
,
axes
=
[
1
],
starts
=
[
index
],
ends
=
[
index
+
1
])
self
.
input
=
fluid
.
layers
.
reshape
(
self
.
_
input
=
fluid
.
layers
.
reshape
(
self
.
input
,
shape
=
[
-
1
,
self
.
_hidden_size
])
self
.
_
input
,
shape
=
[
-
1
,
self
.
_hidden_size
])
for
k
in
range
(
self
.
_num_layers
):
for
k
in
range
(
self
.
_num_layers
):
pre_hidden
=
self
.
hidden_array
[
k
]
pre_hidden
=
self
.
hidden_array
[
k
]
print
(
"pre_hidden shape is:{}"
.
format
(
pre_hidden
.
shape
))
print
(
"input shape is:{}"
.
format
(
self
.
input
.
shape
))
pre_cell
=
self
.
cell_array
[
k
]
pre_cell
=
self
.
cell_array
[
k
]
weight_1
=
self
.
weight_1_arr
[
k
]
weight_1
=
self
.
weight_1_arr
[
k
]
bias
=
self
.
bias_arr
[
k
]
bias
=
self
.
bias_arr
[
k
]
nn
=
fluid
.
layers
.
concat
([
self
.
input
,
pre_hidden
],
1
)
nn
=
fluid
.
layers
.
concat
([
self
.
_
input
,
pre_hidden
],
1
)
gate_input
=
fluid
.
layers
.
matmul
(
x
=
nn
,
y
=
weight_1
)
gate_input
=
fluid
.
layers
.
matmul
(
x
=
nn
,
y
=
weight_1
)
gate_input
=
fluid
.
layers
.
elementwise_add
(
gate_input
,
bias
)
gate_input
=
fluid
.
layers
.
elementwise_add
(
gate_input
,
bias
)
print
(
"gate_input shape is: {}"
.
format
(
gate_input
.
shape
))
i
,
j
,
f
,
o
=
fluid
.
layers
.
split
(
print
(
"gate_input value is :{}"
.
format
(
gate_input
.
_numpy
()))
gate_input
,
num_or_sections
=
4
,
dim
=-
1
)
print
(
"gate_input desc is :{}"
.
format
(
gate_input
))
c
=
pre_cell
*
fluid
.
layers
.
sigmoid
(
f
)
+
fluid
.
layers
.
sigmoid
(
# i, j, f, o = fluid.layers.split(gate_input, num_or_sections=4, dim=-1)
i
)
*
fluid
.
layers
.
tanh
(
j
)
# #
m
=
fluid
.
layers
.
tanh
(
c
)
*
fluid
.
layers
.
sigmoid
(
o
)
# # c = pre_cell * fluid.layers.sigmoid(f) + fluid.layers.sigmoid(
self
.
hidden_array
[
k
]
=
m
# # i) * fluid.layers.tanh(j)
self
.
cell_array
[
k
]
=
c
# # m = fluid.layers.tanh(c) * fluid.layers.sigmoid(o)
self
.
_input
=
m
# #
# # self.hidden_array[k] = m
if
self
.
_dropout
is
not
None
and
self
.
_dropout
>
0.0
:
# # self.cell_array[k] = c
self
.
_input
=
fluid
.
layers
.
dropout
(
# # self.input = m
self
.
_input
,
# #
dropout_prob
=
self
.
_dropout
,
# # if self.dropout is not None and self.dropout > 0.0:
dropout_implementation
=
'upscale_in_train'
)
# # self.input = fluid.layers.dropout(
res
.
append
(
# # self.input,
fluid
.
layers
.
reshape
(
# # dropout_prob=self.dropout,
self
.
_input
,
shape
=
[
1
,
-
1
,
self
.
_hidden_size
]))
# # dropout_implementation='upscale_in_train')
real_res
=
fluid
.
layers
.
concat
(
res
,
0
)
# #
real_res
=
fluid
.
layers
.
transpose
(
x
=
real_res
,
perm
=
[
1
,
0
,
2
])
# # res.append(
last_hidden
=
fluid
.
layers
.
concat
(
self
.
hidden_array
,
1
)
# # fluid.layers.reshape(
last_hidden
=
fluid
.
layers
.
reshape
(
# # input, shape=[1, -1, self._hidden_size]))
last_hidden
,
shape
=
[
-
1
,
self
.
_num_layers
,
self
.
_hidden_size
])
# # real_res = fluid.layers.concat(res, 0)
last_hidden
=
fluid
.
layers
.
transpose
(
x
=
last_hidden
,
perm
=
[
1
,
0
,
2
])
# # real_res = fluid.layers.transpose(x=real_res, perm=[1, 0, 2])
last_cell
=
fluid
.
layers
.
concat
(
self
.
cell_array
,
1
)
# # last_hidden = fluid.layers.concat(self.hidden_array, 1)
last_cell
=
fluid
.
layers
.
reshape
(
# # last_hidden = fluid.layers.reshape(
last_cell
,
shape
=
[
-
1
,
self
.
_num_layers
,
self
.
_hidden_size
])
# # last_hidden, shape=[-1, self._num_layers, self._hidden_size])
last_cell
=
fluid
.
layers
.
transpose
(
x
=
last_cell
,
perm
=
[
1
,
0
,
2
])
# # last_hidden = fluid.layers.transpose(x=last_hidden, perm=[1, 0, 2])
return
real_res
,
last_hidden
,
last_cell
# # last_cell = fluid.layers.concat(self.cell_array, 1)
# # last_cell = fluid.layers.reshape(
# # last_cell, shape=[-1, self._num_layers, self._hidden_size])
# # last_cell = fluid.layers.transpose(x=last_cell, perm=[1, 0, 2])
# #
# return real_res, last_hidden, last_cell
return
[
1
],
[
2
],
[
3
]
class
PtbModel
(
fluid
.
imperative
.
Layer
):
class
PtbModel
(
fluid
.
imperative
.
Layer
):
...
@@ -189,12 +182,11 @@ class PtbModel(fluid.imperative.Layer):
...
@@ -189,12 +182,11 @@ class PtbModel(fluid.imperative.Layer):
x_emb
,
x_emb
,
dropout_prob
=
self
.
drop_out
,
dropout_prob
=
self
.
drop_out
,
dropout_implementation
=
'upscale_in_train'
)
dropout_implementation
=
'upscale_in_train'
)
print
(
"init_c is {}"
.
format
(
init_c
))
rnn_out
,
last_hidden
,
last_cell
=
self
.
simple_lstm_rnn
(
x_emb
,
init_h
,
rnn_out
,
last_hidden
,
last_cell
=
self
.
simple_lstm_rnn
(
x_emb
,
init_h
,
init_c
)
init_c
)
rnn_out
=
fluid
.
layers
.
reshape
(
rnn_out
=
fluid
.
layers
.
reshape
(
rnn_out
,
shape
=
[
-
1
,
self
.
num_steps
,
self
.
hidden_size
])
rnn_out
,
shape
=
[
-
1
,
self
.
num_steps
,
self
.
hidden_size
])
projection
=
fluid
.
layers
.
reshape
(
rnn_out
,
self
.
softmax_weight
)
projection
=
fluid
.
layers
.
matmul
(
rnn_out
,
self
.
softmax_weight
)
projection
=
fluid
.
layers
.
elementwise_add
(
projection
,
self
.
softmax_bias
)
projection
=
fluid
.
layers
.
elementwise_add
(
projection
,
self
.
softmax_bias
)
projection
=
fluid
.
layers
.
reshape
(
projection
=
fluid
.
layers
.
reshape
(
projection
,
shape
=
[
-
1
,
self
.
vocab_size
])
projection
,
shape
=
[
-
1
,
self
.
vocab_size
])
...
@@ -232,7 +224,8 @@ class TestImperativePtbRnn(unittest.TestCase):
...
@@ -232,7 +224,8 @@ class TestImperativePtbRnn(unittest.TestCase):
init_scale
=
init_scale
)
init_scale
=
init_scale
)
sgd
=
SGDOptimizer
(
learning_rate
=
1e-3
)
sgd
=
SGDOptimizer
(
learning_rate
=
1e-3
)
print
(
"q"
)
dy_param_updated
=
dict
()
dy_param_init
=
dict
()
for
i
in
range
(
2
):
for
i
in
range
(
2
):
x_data
=
np
.
arange
(
12
).
reshape
(
4
,
3
).
astype
(
'int64'
)
x_data
=
np
.
arange
(
12
).
reshape
(
4
,
3
).
astype
(
'int64'
)
y_data
=
np
.
arange
(
1
,
13
).
reshape
(
4
,
3
).
astype
(
'int64'
)
y_data
=
np
.
arange
(
1
,
13
).
reshape
(
4
,
3
).
astype
(
'int64'
)
...
@@ -248,17 +241,95 @@ class TestImperativePtbRnn(unittest.TestCase):
...
@@ -248,17 +241,95 @@ class TestImperativePtbRnn(unittest.TestCase):
init_cell
=
to_variable
(
init_cell_data
)
init_cell
=
to_variable
(
init_cell_data
)
dy_loss
,
last_hidden
,
last_cell
=
ptb_model
(
x
,
y
,
init_hidden
,
dy_loss
,
last_hidden
,
last_cell
=
ptb_model
(
x
,
y
,
init_hidden
,
init_cell
)
init_cell
)
dy_param_init
=
dict
()
if
i
==
0
:
if
i
==
0
:
for
param
in
fluid
.
default_main_program
().
global_block
(
for
param
in
fluid
.
default_main_program
().
global_block
(
).
all_parameters
():
).
all_parameters
():
dy_param_init
[
param
.
name
]
=
param
.
_numpy
()
dy_param_init
[
param
.
name
]
=
param
.
_numpy
()
dy_loss
.
_backward
()
dy_loss
.
_backward
()
sgd
.
minimize
(
dy_loss
)
sgd
.
minimize
(
dy_loss
)
dy_param_updated
=
dict
()
for
param
in
fluid
.
default_main_program
().
global_block
(
for
param
in
fluid
.
default_main_program
().
global_block
(
).
all_parameters
():
).
all_parameters
():
dy_param_updated
[
param
.
name
]
=
param
.
_numpy
()
dy_param_updated
[
param
.
name
]
=
param
.
_numpy
()
# print("dy_loss is {}".format(dy_loss._numpy()))
# print("last_hidden is {}".format(last_hidden._numpy()))
# print("last_cell is {}".format(last_cell._numpy()))
with
new_program_scope
():
fluid
.
default_startup_program
().
random_seed
=
seed
fluid
.
default_main_program
().
random_seed
=
seed
# TODO: marsyang1993 Change seed to
ptb_model
=
PtbModel
(
hidden_size
=
hidden_size
,
vocab_size
=
vocab_size
,
num_layers
=
num_layers
,
num_steps
=
num_steps
,
init_scale
=
init_scale
)
exe
=
fluid
.
Executor
(
fluid
.
CPUPlace
())
sgd
=
SGDOptimizer
(
learning_rate
=
1e-3
)
x
=
fluid
.
layers
.
data
(
name
=
"x"
,
shape
=
[
-
1
,
3
,
1
],
dtype
=
'int64'
)
y
=
fluid
.
layers
.
data
(
name
=
"y"
,
shape
=
[
-
1
,
1
],
dtype
=
'float32'
)
init_hidden
=
fluid
.
layers
.
data
(
name
=
"init_hidden"
,
shape
=
[
1
],
dtype
=
'float32'
)
init_cell
=
fluid
.
layers
.
data
(
name
=
"init_cell"
,
shape
=
[
1
],
dtype
=
'float32'
)
static_loss
,
static_last_hidden
,
static_last_cell
=
ptb_model
(
x
,
y
,
init_hidden
,
init_cell
)
sgd
.
minimize
(
static_loss
)
static_param_updated
=
dict
()
static_param_init
=
dict
()
static_param_name_list
=
list
()
for
param
in
fluid
.
default_startup_program
().
global_block
(
).
all_parameters
():
static_param_name_list
.
append
(
param
.
name
)
out
=
exe
.
run
(
framework
.
default_startup_program
(),
fetch_list
=
static_param_name_list
)
for
i
in
range
(
len
(
static_param_name_list
)):
static_param_init
[
static_param_name_list
[
i
]]
=
out
[
i
]
for
i
in
range
(
2
):
x_data
=
np
.
arange
(
12
).
reshape
(
4
,
3
).
astype
(
'int64'
)
y_data
=
np
.
arange
(
1
,
13
).
reshape
(
4
,
3
).
astype
(
'int64'
)
x_data
=
x_data
.
reshape
((
-
1
,
num_steps
,
1
))
y_data
=
y_data
.
reshape
((
-
1
,
1
))
init_hidden_data
=
np
.
zeros
(
(
num_layers
,
batch_size
,
hidden_size
),
dtype
=
'float32'
)
init_cell_data
=
np
.
zeros
(
(
num_layers
,
batch_size
,
hidden_size
),
dtype
=
'float32'
)
fetch_list
=
[
static_loss
,
static_last_hidden
,
static_last_cell
]
fetch_list
.
extend
(
static_param_name_list
)
out
=
exe
.
run
(
fluid
.
default_main_program
(),
feed
=
{
"x"
:
x_data
,
"y"
:
y_data
,
"init_hidden"
:
init_hidden_data
,
"init_cell"
:
init_cell_data
},
fetch_list
=
fetch_list
)
static_loss_value
=
out
[
0
]
static_last_cell_value
=
out
[
1
]
static_last_hidden_value
=
out
[
2
]
# print("static_loss is {}".format(out[0]))
# print("last_hidden is {}".format(out[1]))
# print("last_cell is {}".format(out[2]))
for
i
in
range
(
3
,
len
(
out
)):
static_param_updated
[
static_param_name_list
[
i
-
3
]]
=
out
[
i
]
self
.
assertTrue
(
np
.
allclose
(
static_loss_value
.
all
(),
dy_loss
.
_numpy
().
all
()))
self
.
assertTrue
(
np
.
allclose
(
static_last_cell_value
.
all
(),
last_cell
.
_numpy
().
all
()))
self
.
assertTrue
(
np
.
allclose
(
static_last_hidden_value
.
all
(),
last_hidden
.
_numpy
().
all
()))
for
key
,
value
in
six
.
iteritems
(
static_param_init
):
self
.
assertTrue
(
np
.
allclose
(
value
.
all
(),
dy_param_init
[
key
].
all
()))
for
key
,
value
in
six
.
iteritems
(
static_param_updated
):
self
.
assertTrue
(
np
.
allclose
(
value
.
all
(),
dy_param_updated
[
key
].
all
()))
if
__name__
==
'__main__'
:
if
__name__
==
'__main__'
:
...
...
python/paddle/fluid/tests/unittests/test_imperative_split.py
浏览文件 @
f364b722
...
@@ -38,7 +38,6 @@ class TestImperativePtbRnn(unittest.TestCase):
...
@@ -38,7 +38,6 @@ class TestImperativePtbRnn(unittest.TestCase):
inp
=
to_variable
(
np
.
arange
(
160
).
reshape
(
4
,
40
).
astype
(
'float32'
))
inp
=
to_variable
(
np
.
arange
(
160
).
reshape
(
4
,
40
).
astype
(
'float32'
))
st
=
Split_test
()
st
=
Split_test
()
out
=
st
(
inp
)
out
=
st
(
inp
)
print
(
out
)
if
__name__
==
'__main__'
:
if
__name__
==
'__main__'
:
...
...
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