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
夏有凉风,冬有雪
Paddle
提交
477d92bc
P
Paddle
项目概览
夏有凉风,冬有雪
/
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
477d92bc
编写于
9月 11, 2017
作者:
Y
Yibing Liu
浏览文件
操作
浏览文件
下载
差异文件
merge conflicts
上级
dd64349a
4fbc03d3
变更
21
隐藏空白更改
内联
并排
Showing
21 changed file
with
535 addition
and
66 deletion
+535
-66
paddle/framework/operator.cc
paddle/framework/operator.cc
+9
-0
paddle/framework/operator.h
paddle/framework/operator.h
+5
-2
paddle/gserver/layers/DetectionOutputLayer.cpp
paddle/gserver/layers/DetectionOutputLayer.cpp
+7
-1
paddle/gserver/layers/DetectionUtil.cpp
paddle/gserver/layers/DetectionUtil.cpp
+1
-1
paddle/gserver/layers/DetectionUtil.h
paddle/gserver/layers/DetectionUtil.h
+1
-1
paddle/operators/sum_op.cc
paddle/operators/sum_op.cc
+73
-0
paddle/operators/sum_op.cu
paddle/operators/sum_op.cu
+18
-0
paddle/operators/sum_op.h
paddle/operators/sum_op.h
+65
-0
paddle/platform/enforce.h
paddle/platform/enforce.h
+1
-18
paddle/pybind/pybind.cc
paddle/pybind/pybind.cc
+4
-0
paddle/scripts/docker/build.sh
paddle/scripts/docker/build.sh
+4
-0
python/paddle/trainer/config_parser.py
python/paddle/trainer/config_parser.py
+2
-2
python/paddle/trainer_config_helpers/layers.py
python/paddle/trainer_config_helpers/layers.py
+3
-1
python/paddle/v2/event.py
python/paddle/v2/event.py
+8
-2
python/paddle/v2/framework/op.py
python/paddle/v2/framework/op.py
+8
-2
python/paddle/v2/framework/tests/CMakeLists.txt
python/paddle/v2/framework/tests/CMakeLists.txt
+1
-0
python/paddle/v2/framework/tests/op_test.py
python/paddle/v2/framework/tests/op_test.py
+275
-0
python/paddle/v2/framework/tests/test_cross_entropy_op.py
python/paddle/v2/framework/tests/test_cross_entropy_op.py
+9
-18
python/paddle/v2/framework/tests/test_sigmoid_op.py
python/paddle/v2/framework/tests/test_sigmoid_op.py
+10
-16
python/paddle/v2/framework/tests/test_sum_op.py
python/paddle/v2/framework/tests/test_sum_op.py
+24
-0
python/paddle/v2/trainer.py
python/paddle/v2/trainer.py
+7
-2
未找到文件。
paddle/framework/operator.cc
浏览文件 @
477d92bc
...
...
@@ -123,6 +123,15 @@ OperatorBase::OperatorBase(const std::string& type,
CheckAllInputOutputSet
();
}
std
::
vector
<
std
::
string
>
OperatorBase
::
InputVars
()
const
{
std
::
vector
<
std
::
string
>
ret_val
;
for
(
auto
&
o
:
outputs_
)
{
ret_val
.
reserve
(
ret_val
.
size
()
+
o
.
second
.
size
());
ret_val
.
insert
(
ret_val
.
end
(),
o
.
second
.
begin
(),
o
.
second
.
end
());
}
return
ret_val
;
}
std
::
vector
<
std
::
string
>
OperatorBase
::
OutputVars
(
bool
has_intermediate
)
const
{
std
::
vector
<
std
::
string
>
ret_val
;
if
(
has_intermediate
)
{
...
...
paddle/framework/operator.h
浏览文件 @
477d92bc
...
...
@@ -94,11 +94,14 @@ class OperatorBase {
const
VariableNameMap
&
Inputs
()
const
{
return
inputs_
;
}
const
VariableNameMap
&
Outputs
()
const
{
return
outputs_
;
}
//! Get a input with argument's name described in `op_proto`
std
::
string
Input
(
const
std
::
string
&
name
)
const
;
//! Get a input which has multiple variables.
const
std
::
vector
<
std
::
string
>&
Inputs
(
const
std
::
string
&
name
)
const
;
std
::
vector
<
std
::
string
>
InputVars
()
const
;
//! Get a output with argument's name described in `op_proto`
std
::
string
Output
(
const
std
::
string
&
name
)
const
;
//! Get an output which has multiple variables.
...
...
@@ -311,9 +314,9 @@ class InferShapeContext {
}
template
<
typename
T
>
std
::
vector
<
const
T
*>
MultiOutput
(
const
std
::
string
&
name
)
const
{
std
::
vector
<
T
*>
MultiOutput
(
const
std
::
string
&
name
)
const
{
auto
names
=
op_
.
Outputs
(
name
);
std
::
vector
<
const
T
*>
res
;
std
::
vector
<
T
*>
res
;
res
.
reserve
(
names
.
size
());
std
::
transform
(
names
.
begin
(),
names
.
end
(),
std
::
back_inserter
(
res
),
[
&
](
const
std
::
string
&
sub_name
)
{
...
...
paddle/gserver/layers/DetectionOutputLayer.cpp
浏览文件 @
477d92bc
...
...
@@ -139,7 +139,13 @@ void DetectionOutputLayer::forward(PassType passType) {
allDecodedBBoxes
,
&
allIndices
);
resetOutput
(
numKept
,
7
);
if
(
numKept
>
0
)
{
resetOutput
(
numKept
,
7
);
}
else
{
MatrixPtr
outV
=
getOutputValue
();
outV
=
NULL
;
return
;
}
MatrixPtr
outV
=
getOutputValue
();
getDetectionOutput
(
confBuffer_
->
getData
(),
numKept
,
...
...
paddle/gserver/layers/DetectionUtil.cpp
浏览文件 @
477d92bc
...
...
@@ -469,7 +469,7 @@ size_t getDetectionIndices(
const
size_t
numClasses
,
const
size_t
backgroundId
,
const
size_t
batchSize
,
const
size_t
confThreshold
,
const
real
confThreshold
,
const
size_t
nmsTopK
,
const
real
nmsThreshold
,
const
size_t
keepTopK
,
...
...
paddle/gserver/layers/DetectionUtil.h
浏览文件 @
477d92bc
...
...
@@ -275,7 +275,7 @@ size_t getDetectionIndices(
const
size_t
numClasses
,
const
size_t
backgroundId
,
const
size_t
batchSize
,
const
size_t
confThreshold
,
const
real
confThreshold
,
const
size_t
nmsTopK
,
const
real
nmsThreshold
,
const
size_t
keepTopK
,
...
...
paddle/operators/sum_op.cc
0 → 100644
浏览文件 @
477d92bc
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
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. */
#include "paddle/operators/sum_op.h"
#include <vector>
namespace
paddle
{
namespace
operators
{
using
framework
::
Tensor
;
class
SumOp
:
public
framework
::
OperatorWithKernel
{
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
auto
ins
=
ctx
.
MultiInput
<
framework
::
Tensor
>
(
"X"
);
auto
*
out
=
ctx
.
Output
<
framework
::
Tensor
>
(
"Out"
);
int
N
=
ins
.
size
();
auto
in_dim
=
ins
[
0
]
->
dims
();
PADDLE_ENFORCE_GT
(
N
,
1
,
"Input tensors count should > 1."
);
for
(
int
i
=
1
;
i
<
N
;
i
++
)
{
auto
dim
=
ins
[
i
]
->
dims
();
PADDLE_ENFORCE
(
in_dim
==
dim
,
"Input tensors must have same shape"
);
}
out
->
Resize
(
in_dim
);
}
};
class
SumOpMaker
:
public
framework
::
OpProtoAndCheckerMaker
{
public:
SumOpMaker
(
framework
::
OpProto
*
proto
,
framework
::
OpAttrChecker
*
op_checker
)
:
OpProtoAndCheckerMaker
(
proto
,
op_checker
)
{
AddInput
(
"X"
,
"the input tensors of sum operator."
).
AsDuplicable
();
AddOutput
(
"Out"
,
"the output tensor of sum operator."
);
AddComment
(
R"DOC(
Sum the input tensors.
)DOC"
);
}
};
class
SumGradOp
:
public
framework
::
OperatorWithKernel
{
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
protected:
void
InferShape
(
const
framework
::
InferShapeContext
&
ctx
)
const
override
{
auto
outputs
=
ctx
.
MultiOutput
<
Tensor
>
(
framework
::
GradVarName
(
"X"
));
auto
dims
=
ctx
.
Input
<
Tensor
>
(
framework
::
GradVarName
(
"Out"
))
->
dims
();
for
(
auto
output
:
outputs
)
{
output
->
Resize
(
dims
);
}
}
};
}
// namespace operators
}
// namespace paddle
namespace
ops
=
paddle
::
operators
;
REGISTER_OP
(
sum
,
ops
::
SumOp
,
ops
::
SumOpMaker
,
sum_grad
,
ops
::
SumGradOp
);
REGISTER_OP_CPU_KERNEL
(
sum
,
ops
::
SumKernel
<
paddle
::
platform
::
CPUPlace
,
float
>
);
REGISTER_OP_CPU_KERNEL
(
sum_grad
,
ops
::
SumGradKernel
<
paddle
::
platform
::
CPUPlace
,
float
>
);
paddle/operators/sum_op.cu
0 → 100644
浏览文件 @
477d92bc
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
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. */
#define EIGEN_USE_GPU
#include "paddle/operators/sum_op.h"
namespace
ops
=
paddle
::
operators
;
REGISTER_OP_GPU_KERNEL
(
sum
,
ops
::
SumKernel
<
paddle
::
platform
::
GPUPlace
,
float
>
);
REGISTER_OP_GPU_KERNEL
(
sum_grad
,
ops
::
SumGradKernel
<
paddle
::
platform
::
GPUPlace
,
float
>
);
paddle/operators/sum_op.h
0 → 100644
浏览文件 @
477d92bc
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
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. */
#pragma once
#include "paddle/framework/eigen.h"
#include "paddle/framework/op_registry.h"
namespace
paddle
{
namespace
operators
{
using
Tensor
=
framework
::
Tensor
;
template
<
typename
T
,
int
MajorType
=
Eigen
::
RowMajor
,
typename
IndexType
=
Eigen
::
DenseIndex
>
using
EigenVector
=
framework
::
EigenVector
<
T
,
MajorType
,
IndexType
>
;
template
<
typename
Place
,
typename
T
>
class
SumKernel
:
public
framework
::
OpKernel
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
context
)
const
override
{
auto
ins
=
context
.
MultiInput
<
Tensor
>
(
"X"
);
auto
*
out
=
context
.
Output
<
Tensor
>
(
"Out"
);
out
->
mutable_data
<
T
>
(
context
.
GetPlace
());
auto
place
=
context
.
GetEigenDevice
<
Place
>
();
auto
result
=
EigenVector
<
T
>::
Flatten
(
*
out
);
int
N
=
ins
.
size
();
auto
in
=
EigenVector
<
T
>::
Flatten
(
*
(
ins
[
0
]));
result
.
device
(
place
)
=
in
;
for
(
int
i
=
1
;
i
<
N
;
i
++
)
{
auto
in
=
EigenVector
<
T
>::
Flatten
(
*
(
ins
[
i
]));
result
.
device
(
place
)
=
result
+
in
;
}
}
};
template
<
typename
Place
,
typename
T
>
class
SumGradKernel
:
public
framework
::
OpKernel
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
context
)
const
override
{
auto
*
input
=
context
.
Input
<
Tensor
>
(
framework
::
GradVarName
(
"Out"
));
auto
outs
=
context
.
MultiOutput
<
Tensor
>
(
framework
::
GradVarName
(
"X"
));
for
(
auto
out
:
outs
)
{
out
->
mutable_data
<
T
>
(
context
.
GetPlace
());
}
auto
place
=
context
.
GetEigenDevice
<
Place
>
();
auto
in
=
EigenVector
<
T
>::
Flatten
(
*
input
);
for
(
auto
out
:
outs
)
{
auto
result
=
EigenVector
<
T
>::
Flatten
(
*
out
);
result
.
device
(
place
)
=
in
;
}
}
};
}
// namespace operators
}
// namespace paddle
paddle/platform/enforce.h
浏览文件 @
477d92bc
...
...
@@ -25,10 +25,6 @@ limitations under the License. */
#include "paddle/string/printf.h"
#include "paddle/string/to_string.h"
#ifdef __GNUC__
#include <cxxabi.h> // for __cxa_demangle
#endif
#ifndef PADDLE_ONLY_CPU
#include "paddle/platform/dynload/cublas.h"
...
...
@@ -46,19 +42,6 @@ limitations under the License. */
namespace
paddle
{
namespace
platform
{
namespace
{
#ifdef __GNUC__
inline
std
::
string
demangle
(
std
::
string
name
)
{
int
status
=
-
4
;
// some arbitrary value to eliminate the compiler warning
std
::
unique_ptr
<
char
,
void
(
*
)(
void
*
)
>
res
{
abi
::
__cxa_demangle
(
name
.
c_str
(),
NULL
,
NULL
,
&
status
),
std
::
free
};
return
(
status
==
0
)
?
res
.
get
()
:
name
;
}
#else
inline
std
::
string
demangle
(
std
::
string
name
)
{
return
name
;
}
#endif
}
struct
EnforceNotMet
:
public
std
::
exception
{
std
::
exception_ptr
exp_
;
std
::
string
err_str_
;
...
...
@@ -79,7 +62,7 @@ struct EnforceNotMet : public std::exception {
Dl_info
info
;
for
(
int
i
=
0
;
i
<
size
;
++
i
)
{
if
(
dladdr
(
call_stack
[
i
],
&
info
))
{
auto
demangled
=
demangle
(
info
.
dli_sname
)
;
auto
demangled
=
info
.
dli_sname
;
auto
addr_offset
=
static_cast
<
char
*>
(
call_stack
[
i
])
-
static_cast
<
char
*>
(
info
.
dli_saddr
);
sout
<<
string
::
Sprintf
(
"%-3d %*0p %s + %zd
\n
"
,
i
,
...
...
paddle/pybind/pybind.cc
浏览文件 @
477d92bc
...
...
@@ -51,6 +51,7 @@ USE_CPU_ONLY_OP(gather);
USE_CPU_ONLY_OP
(
scatter
);
USE_OP
(
top_k
);
USE_OP
(
squared_l2_distance
);
USE_OP
(
sum
);
USE_OP
(
reshape
);
namespace
paddle
{
...
...
@@ -217,7 +218,10 @@ All parameter, weight, gradient are variables in Paddle.
->
std
::
map
<
std
::
string
,
std
::
vector
<
std
::
string
>>
{
return
op
.
Outputs
();
})
.
def
(
"output_vars"
,
[](
const
OperatorBase
&
op
)
{
return
op
.
OutputVars
(
true
);
})
.
def
(
"inputs"
,
[](
const
OperatorBase
&
op
)
{
return
op
.
Inputs
();
})
.
def
(
"input_vars"
,
[](
const
OperatorBase
&
op
)
{
return
op
.
InputVars
();
})
.
def
(
"__str__"
,
&
OperatorBase
::
DebugString
)
.
def
(
"no_intermediate_outputs"
,
[](
const
OperatorBase
&
op
)
{
return
op
.
OutputVars
(
false
);
})
...
...
paddle/scripts/docker/build.sh
浏览文件 @
477d92bc
...
...
@@ -30,6 +30,8 @@ Configuring cmake in /paddle/build ...
-DCMAKE_BUILD_TYPE=Release
-DWITH_DOC=OFF
-DWITH_GPU=
${
WITH_GPU
:-
OFF
}
-DWITH_MKLDNN=
${
WITH_MKLDNN
:-
ON
}
-DWITH_MKLML=
${
WITH_MKLML
:-
ON
}
-DWITH_AVX=
${
WITH_AVX
:-
OFF
}
-DWITH_GOLANG=
${
WITH_GOLANG
:-
ON
}
-DWITH_SWIG_PY=ON
...
...
@@ -50,6 +52,8 @@ cmake .. \
-DCMAKE_BUILD_TYPE
=
Release
\
-DWITH_DOC
=
OFF
\
-DWITH_GPU
=
${
WITH_GPU
:-
OFF
}
\
-DWITH_MKLDNN
=
${
WITH_MKLDNN
:-
ON
}
\
-DWITH_MKLML
=
${
WITH_MKLML
:-
ON
}
\
-DWITH_AVX
=
${
WITH_AVX
:-
OFF
}
\
-DWITH_GOLANG
=
${
WITH_GOLANG
:-
ON
}
\
-DWITH_SWIG_PY
=
${
WITH_SWIG_PY
:-
ON
}
\
...
...
python/paddle/trainer/config_parser.py
浏览文件 @
477d92bc
...
...
@@ -3748,8 +3748,8 @@ class SwitchOrderLayer(LayerBase):
def
__init__
(
self
,
name
,
inputs
,
reshape
,
**
xargs
):
super
(
SwitchOrderLayer
,
self
).
__init__
(
name
,
'switch_order'
,
0
,
inputs
=
inputs
,
**
xargs
)
self
.
config
.
reshape_conf
.
height
A
xis
.
extend
(
reshape
[
'height'
])
self
.
config
.
reshape_conf
.
width
A
xis
.
extend
(
reshape
[
'width'
])
self
.
config
.
reshape_conf
.
height
_a
xis
.
extend
(
reshape
[
'height'
])
self
.
config
.
reshape_conf
.
width
_a
xis
.
extend
(
reshape
[
'width'
])
# Deprecated, use a new layer specific class instead
...
...
python/paddle/trainer_config_helpers/layers.py
浏览文件 @
477d92bc
...
...
@@ -1223,7 +1223,8 @@ def detection_output_layer(input_loc,
name
=
None
):
"""
Apply the NMS to the output of network and compute the predict bounding
box location.
box location. The output of this layer could be None if there is no valid
bounding box.
:param name: The Layer Name.
:type name: basestring
...
...
@@ -6460,6 +6461,7 @@ def switch_order_layer(input,
return
LayerOutput
(
name
=
name
,
layer_type
=
LayerType
.
SWITCH_ORDER_LAYER
,
activation
=
act
,
parents
=
input
,
size
=
l
.
config
.
size
)
...
...
python/paddle/v2/event.py
浏览文件 @
477d92bc
...
...
@@ -53,10 +53,13 @@ class BeginPass(object):
class
EndPass
(
WithMetric
):
"""
Event On One Pass Training Complete.
To get the output of a specific layer, add "event.gm.getLayerOutputs('predict_layer')"
in your event_handler call back
"""
def
__init__
(
self
,
pass_id
,
evaluator
):
def
__init__
(
self
,
pass_id
,
evaluator
,
gm
):
self
.
pass_id
=
pass_id
self
.
gm
=
gm
WithMetric
.
__init__
(
self
,
evaluator
)
...
...
@@ -73,10 +76,13 @@ class BeginIteration(object):
class
EndIteration
(
WithMetric
):
"""
Event On One Batch Training Complete.
To get the output of a specific layer, add "event.gm.getLayerOutputs('predict_layer')"
in your event_handler call back
"""
def
__init__
(
self
,
pass_id
,
batch_id
,
cost
,
evaluator
):
def
__init__
(
self
,
pass_id
,
batch_id
,
cost
,
evaluator
,
gm
):
self
.
pass_id
=
pass_id
self
.
batch_id
=
batch_id
self
.
cost
=
cost
self
.
gm
=
gm
WithMetric
.
__init__
(
self
,
evaluator
)
python/paddle/v2/framework/op.py
浏览文件 @
477d92bc
...
...
@@ -142,8 +142,8 @@ def create_op_creation_method(op_proto):
return
OpInfo
(
method
=
__impl__
,
name
=
op_proto
.
type
,
inputs
=
[
var
.
name
for
var
in
op_proto
.
inputs
],
outputs
=
[
var
.
name
for
var
in
op_proto
.
outputs
],
inputs
=
[
(
var
.
name
,
var
.
duplicable
)
for
var
in
op_proto
.
inputs
],
outputs
=
[
(
var
.
name
,
var
.
duplicable
)
for
var
in
op_proto
.
outputs
],
attrs
=
[
attr
.
name
for
attr
in
op_proto
.
attrs
])
...
...
@@ -180,9 +180,15 @@ class OperatorFactory(object):
return
self
.
op_methods
.
get
(
t
)
def
get_op_input_names
(
self
,
type
):
return
map
(
lambda
x
:
x
[
0
],
self
.
get_op_info
(
type
).
inputs
)
def
get_op_inputs
(
self
,
type
):
return
self
.
get_op_info
(
type
).
inputs
def
get_op_output_names
(
self
,
type
):
return
map
(
lambda
x
:
x
[
0
],
self
.
get_op_info
(
type
).
outputs
)
def
get_op_outputs
(
self
,
type
):
return
self
.
get_op_info
(
type
).
outputs
def
get_op_attr_names
(
self
,
type
):
...
...
python/paddle/v2/framework/tests/CMakeLists.txt
浏览文件 @
477d92bc
...
...
@@ -33,6 +33,7 @@ py_test(test_sgd_op SRCS test_sgd_op.py)
py_test
(
test_gradient_checker SRCS test_gradient_checker.py
)
py_test
(
test_lookup_table SRCS test_lookup_table.py
)
py_test
(
test_scale_and_identity_op SRCS test_scale_and_identity_op.py
)
py_test
(
test_sum_op SRCS test_sum_op.py
)
py_test
(
mnist SRCS mnist.py
)
py_test
(
test_squared_l2_distance_op SRCS test_squared_l2_distance_op.py
)
py_test
(
test_reshape_op SRCS test_reshape_op.py
)
python/paddle/v2/framework/tests/op_test.py
0 → 100644
浏览文件 @
477d92bc
import
unittest
import
numpy
as
np
import
itertools
import
paddle.v2.framework.core
as
core
from
paddle.v2.framework.op
import
Operator
def
grad_var_name
(
var_name
):
return
var_name
+
"@GRAD"
def
create_op
(
scope
,
op_type
,
inputs
,
outputs
,
attrs
=
None
):
kwargs
=
dict
()
for
in_name
,
in_dup
in
Operator
.
get_op_inputs
(
op_type
):
if
in_name
in
inputs
:
kwargs
[
in_name
]
=
[]
if
in_dup
:
sub_in
=
inputs
[
in_name
]
for
sub_in_name
in
sub_in
:
var
=
scope
.
new_var
(
sub_in_name
)
kwargs
[
in_name
].
append
(
sub_in_name
)
else
:
var
=
scope
.
new_var
(
in_name
)
kwargs
[
in_name
].
append
(
in_name
)
for
out_name
,
out_dup
in
Operator
.
get_op_outputs
(
op_type
):
if
out_name
in
outputs
:
kwargs
[
out_name
]
=
[]
if
out_dup
:
sub_in
=
outputs
[
out_name
]
for
sun_in_name
in
sub_in
:
var
=
scope
.
new_var
(
sun_in_name
)
kwargs
[
out_name
].
append
(
sun_in_name
)
else
:
var
=
scope
.
new_var
(
out_name
)
kwargs
[
out_name
].
append
(
out_name
)
for
attr_name
in
Operator
.
get_op_attr_names
(
op_type
):
kwargs
[
attr_name
]
=
attrs
[
attr_name
]
return
Operator
(
op_type
,
**
kwargs
)
def
set_input
(
scope
,
op
,
inputs
,
place
):
for
in_name
,
in_dup
in
Operator
.
get_op_inputs
(
op
.
type
()):
if
in_name
in
inputs
:
if
in_dup
:
sub_in
=
inputs
[
in_name
]
for
sub_in_name
in
sub_in
:
var
=
scope
.
find_var
(
sub_in_name
)
tensor
=
var
.
get_tensor
()
arr
=
sub_in
[
sub_in_name
]
tensor
.
set_dims
(
arr
.
shape
)
tensor
.
set
(
arr
,
place
)
else
:
var
=
scope
.
find_var
(
in_name
)
tensor
=
var
.
get_tensor
()
arr
=
inputs
[
in_name
]
tensor
.
set_dims
(
arr
.
shape
)
tensor
.
set
(
arr
,
place
)
def
set_output_grad
(
scope
,
op
,
outputs
,
place
):
for
out_name
,
out_dup
in
Operator
.
get_op_outputs
(
op
.
type
()):
if
out_name
in
outputs
:
if
out_dup
:
sub_out
=
outputs
[
out_name
]
for
sub_out_name
in
sub_out
:
out_tensor
=
scope
.
find_var
(
sub_out_name
).
get_tensor
()
grad_tensor
=
scope
.
new_var
(
grad_var_name
(
sub_out_name
)).
get_tensor
()
grad_tensor
.
set_dims
(
out_tensor
.
shape
())
data
=
np
.
ones
(
out_tensor
.
shape
(),
dtype
=
np
.
float32
)
grad_tensor
.
set
(
data
,
place
)
else
:
out_tensor
=
scope
.
find_var
(
out_name
).
get_tensor
()
grad_tensor
=
scope
.
new_var
(
grad_var_name
(
out_name
)).
get_tensor
(
)
grad_tensor
.
set_dims
(
out_tensor
.
shape
())
data
=
np
.
ones
(
out_tensor
.
shape
(),
dtype
=
np
.
float32
)
grad_tensor
.
set
(
data
,
place
)
def
get_numeric_gradient
(
scope
,
op
,
inputs
,
input_to_check
,
output_name
,
delta
=
0.005
,
in_place
=
False
):
set_input
(
scope
,
op
,
inputs
,
core
.
CPUPlace
())
op
.
infer_shape
(
scope
)
tensor_to_check
=
scope
.
find_var
(
input_to_check
).
get_tensor
()
def
product
(
dim
):
return
reduce
(
lambda
a
,
b
:
a
*
b
,
dim
,
1
)
ctx
=
core
.
DeviceContext
.
create
(
core
.
CPUPlace
())
def
get_output
():
op
.
run
(
scope
,
ctx
)
return
np
.
array
(
scope
.
find_var
(
output_name
).
get_tensor
()).
sum
()
tensor_to_check
=
scope
.
find_var
(
input_to_check
).
get_tensor
()
tensor_size
=
product
(
tensor_to_check
.
get_dims
())
gradient_flat
=
np
.
zeros
(
shape
=
(
tensor_size
,
),
dtype
=
'float32'
)
# we only compute gradient of one element each time.
# we use a for loop to compute the gradient of every element.
for
i
in
xrange
(
tensor_size
):
if
in_place
:
set_input
(
op
,
inputs
,
core
.
CPUPlace
())
# get one input element throw it's index i.
origin
=
tensor_to_check
.
get_float_element
(
i
)
# add delta to it, run op and then get the sum of the result tensor.
x_pos
=
origin
+
delta
tensor_to_check
.
set_float_element
(
i
,
x_pos
)
y_pos
=
get_output
()
if
in_place
:
set_input
(
op
,
inputs
,
core
.
CPUPlace
())
x_neg
=
origin
-
delta
tensor_to_check
.
set_float_element
(
i
,
x_neg
)
y_neg
=
get_output
()
tensor_to_check
.
set_float_element
(
i
,
origin
)
gradient_flat
[
i
]
=
(
y_pos
-
y_neg
)
/
delta
/
2
return
gradient_flat
.
reshape
(
tensor_to_check
.
get_dims
())
def
get_backward_op
(
scope
,
op
,
no_grad_set
):
backward_op
=
core
.
Operator
.
backward
(
op
,
no_grad_set
)
for
input
in
backward_op
.
input_vars
():
var
=
scope
.
new_var
(
input
)
var
.
get_tensor
()
for
output
in
backward_op
.
output_vars
():
var
=
scope
.
new_var
(
output
)
var
.
get_tensor
()
return
backward_op
def
get_gradient
(
scope
,
op
,
inputs
,
outputs
,
grad_name
,
place
,
no_grad_set
=
None
):
ctx
=
core
.
DeviceContext
.
create
(
place
)
set_input
(
scope
,
op
,
inputs
,
place
)
op
.
infer_shape
(
scope
)
op
.
run
(
scope
,
ctx
)
if
no_grad_set
is
None
:
no_grad_set
=
set
()
backward_op
=
get_backward_op
(
scope
,
op
,
no_grad_set
)
set_output_grad
(
scope
,
op
,
outputs
,
place
)
backward_op
.
infer_shape
(
scope
)
backward_op
.
run
(
scope
,
ctx
)
out
=
np
.
array
(
scope
.
find_var
(
grad_name
).
get_tensor
())
return
out
class
OpTest
(
unittest
.
TestCase
):
def
check_output_with_place
(
self
,
place
):
self
.
scope
=
core
.
Scope
()
self
.
op
=
create_op
(
self
.
scope
,
self
.
op_type
,
self
.
inputs
,
self
.
outputs
)
if
isinstance
(
place
,
core
.
GPUPlace
)
and
not
self
.
op
.
support_gpu
():
return
set_input
(
self
.
scope
,
self
.
op
,
self
.
inputs
,
place
)
self
.
op
.
infer_shape
(
self
.
scope
)
ctx
=
core
.
DeviceContext
.
create
(
place
)
self
.
op
.
run
(
self
.
scope
,
ctx
)
for
out_name
,
out_dup
in
Operator
.
get_op_outputs
(
self
.
op
.
type
()):
if
out_dup
:
sub_out
=
self
.
outputs
[
out_name
]
for
sub_out_name
in
sub_out
:
actual
=
np
.
array
(
self
.
scope
.
find_var
(
sub_out_name
).
get_tensor
())
expect
=
sub_out
[
sub_out_name
]
self
.
assertTrue
(
np
.
allclose
(
actual
,
expect
,
atol
=
1e-05
),
"output name: "
+
out_name
+
"has diff"
)
else
:
actual
=
np
.
array
(
self
.
scope
.
find_var
(
out_name
).
get_tensor
())
expect
=
self
.
outputs
[
out_name
]
self
.
assertTrue
(
np
.
allclose
(
actual
,
expect
,
atol
=
1e-05
),
"output name: "
+
out_name
+
"has diff"
)
def
check_output
(
self
):
places
=
[
core
.
CPUPlace
()]
if
core
.
is_compile_gpu
():
places
.
append
(
core
.
GPUPlace
(
0
))
for
place
in
places
:
self
.
check_output_with_place
(
place
)
def
__assert_is_close
(
self
,
numeric_grads
,
analytic_grads
,
names
,
max_relative_error
,
msg_prefix
):
for
a
,
b
,
name
in
itertools
.
izip
(
numeric_grads
,
analytic_grads
,
names
):
abs_a
=
np
.
abs
(
a
)
abs_a
[
abs_a
<
1e-3
]
=
1
diff_mat
=
np
.
abs
(
a
-
b
)
/
abs_a
max_diff
=
np
.
max
(
diff_mat
)
def
err_msg
():
offset
=
np
.
argmax
(
diff_mat
>
max_relative_error
)
return
"%s Variable %s max gradient diff %f over limit %f, the first "
\
"error element is %d"
%
(
msg_prefix
,
name
,
max_diff
,
max_relative_error
,
offset
)
self
.
assertLessEqual
(
max_diff
,
max_relative_error
,
err_msg
())
def
check_grad
(
self
,
inputs_to_check
,
output_name
,
no_grad_set
=
None
,
in_place
=
False
,
max_relative_error
=
0.005
):
self
.
scope
=
core
.
Scope
()
self
.
op
=
create_op
(
self
.
scope
,
self
.
op_type
,
self
.
inputs
,
self
.
outputs
)
if
no_grad_set
is
None
:
no_grad_set
=
set
()
numeric_grads
=
[
get_numeric_gradient
(
self
.
scope
,
self
.
op
,
self
.
inputs
,
input_to_check
,
output_name
,
in_place
=
in_place
)
for
input_to_check
in
inputs_to_check
]
grad_names
=
[
grad_var_name
(
input_to_check
)
for
input_to_check
in
inputs_to_check
]
cpu_place
=
core
.
CPUPlace
()
cpu_analytic_grads
=
[
get_gradient
(
self
.
scope
,
self
.
op
,
self
.
inputs
,
self
.
outputs
,
grad_name
,
cpu_place
,
no_grad_set
)
for
grad_name
in
grad_names
]
self
.
__assert_is_close
(
numeric_grads
,
cpu_analytic_grads
,
grad_names
,
max_relative_error
,
"Gradient Check On %s"
%
str
(
cpu_place
))
if
core
.
is_compile_gpu
()
and
self
.
op
.
support_gpu
():
gpu_place
=
core
.
GPUPlace
(
0
)
gpu_analytic_grads
=
[
get_gradient
(
self
.
scope
,
self
.
op
,
self
.
inputs
,
self
.
outputs
,
grad_name
,
gpu_place
,
no_grad_set
)
for
grad_name
in
grad_names
]
self
.
__assert_is_close
(
numeric_grads
,
gpu_analytic_grads
,
grad_names
,
max_relative_error
,
"Gradient Check On %s"
%
str
(
gpu_place
))
for
c_grad
,
g_grad
,
name
in
itertools
.
izip
(
cpu_analytic_grads
,
gpu_analytic_grads
,
grad_names
):
self
.
assertTrue
(
np
.
allclose
(
c_grad
,
g_grad
,
atol
=
1e-4
),
"output name: "
+
name
+
" has diff"
)
python/paddle/v2/framework/tests/test_cross_entropy_op.py
浏览文件 @
477d92bc
import
unittest
import
numpy
from
op_test_util
import
OpTestMeta
from
gradient_checker
import
GradientChecker
,
create_op
from
op_test
import
OpTest
class
TestCrossEntropy
(
unittest
.
TestCase
):
__metaclass__
=
OpTestMeta
class
TestCrossEntropy
(
OpTest
):
def
setUp
(
self
):
self
.
type
=
"onehot_cross_entropy"
self
.
op_
type
=
"onehot_cross_entropy"
batch_size
=
30
class_num
=
10
X
=
numpy
.
random
.
random
((
batch_size
,
class_num
)).
astype
(
"float32"
)
label
=
5
*
numpy
.
ones
(
batch_size
).
astype
(
"int32"
)
X
=
numpy
.
random
.
uniform
(
0.1
,
1.0
,
[
batch_size
,
class_num
]).
astype
(
"float32"
)
label
=
(
class_num
/
2
)
*
numpy
.
ones
(
batch_size
).
astype
(
"int32"
)
self
.
inputs
=
{
'X'
:
X
,
'label'
:
label
}
Y
=
[]
for
i
in
range
(
0
,
batch_size
):
Y
.
append
(
-
numpy
.
log
(
X
[
i
][
label
[
i
]]))
self
.
outputs
=
{
'Y'
:
numpy
.
array
(
Y
).
astype
(
"float32"
)}
def
test_check_output
(
self
):
self
.
check_output
()
class
CrossEntropyGradOpTest
(
GradientChecker
):
def
test_check_grad
(
self
):
op
=
create_op
(
"onehot_cross_entropy"
)
batch_size
=
30
class_num
=
10
inputs
=
{
"X"
:
numpy
.
random
.
uniform
(
0.1
,
1.0
,
[
batch_size
,
class_num
]).
astype
(
"float32"
),
"label"
:
(
class_num
/
2
)
*
numpy
.
ones
(
batch_size
).
astype
(
"int32"
)
}
self
.
check_grad
(
op
,
inputs
,
set
(
"X"
),
"Y"
)
self
.
check_grad
([
"X"
],
"Y"
)
if
__name__
==
"__main__"
:
...
...
python/paddle/v2/framework/tests/test_sigmoid_op.py
浏览文件 @
477d92bc
import
unittest
import
numpy
as
np
from
op_test_util
import
OpTestMeta
from
gradient_checker
import
GradientChecker
,
create_op
from
op_test
import
OpTest
class
TestSigmoidOp
(
unittest
.
TestCase
):
__metaclass__
=
OpTestMeta
class
TestSigmoid
(
OpTest
):
def
setUp
(
self
):
self
.
type
=
"sigmoid"
self
.
inputs
=
{
'X'
:
np
.
random
.
random
((
15
,
31
)).
astype
(
"float32"
)}
self
.
op_type
=
"sigmoid"
self
.
inputs
=
{
'X'
:
np
.
random
.
uniform
(
0.1
,
1
,
[
11
,
17
]).
astype
(
"float32"
)
}
self
.
outputs
=
{
'Y'
:
1
/
(
1
+
np
.
exp
(
-
self
.
inputs
[
'X'
]))}
def
test_check_output
(
self
):
self
.
check_output
()
class
TestSigmoidGradOp
(
GradientChecker
):
def
test_grad
(
self
):
op
=
create_op
(
"sigmoid"
)
inputs
=
{
"X"
:
np
.
random
.
uniform
(
0.1
,
1
,
[
11
,
17
]).
astype
(
"float32"
)}
# compare gpu and cpu results for backward op.
# this test will be skiped if only compiling CPU version.
self
.
compare_grad
(
op
,
inputs
)
# check gradients
self
.
check_grad
(
op
,
inputs
,
set
(
"X"
),
"Y"
,
max_relative_error
=
0.007
)
def
test_check_grad
(
self
):
self
.
check_grad
([
"X"
],
"Y"
,
max_relative_error
=
0.007
)
if
__name__
==
'__main__'
:
...
...
python/paddle/v2/framework/tests/test_sum_op.py
0 → 100644
浏览文件 @
477d92bc
import
unittest
import
numpy
as
np
from
op_test
import
OpTest
class
TestSumOp
(
OpTest
):
def
setUp
(
self
):
self
.
op_type
=
"sum"
x0
=
np
.
random
.
random
((
3
,
4
)).
astype
(
'float32'
)
x1
=
np
.
random
.
random
((
3
,
4
)).
astype
(
'float32'
)
x2
=
np
.
random
.
random
((
3
,
4
)).
astype
(
'float32'
)
self
.
inputs
=
{
"X"
:
{
"x0"
:
x0
,
"x1"
:
x1
,
"x2"
:
x2
}}
y
=
x0
+
x1
+
x2
self
.
outputs
=
{
'Out'
:
y
}
def
test_check_output
(
self
):
self
.
check_output
()
def
test_check_grad
(
self
):
self
.
check_grad
([
"x0"
],
"Out"
)
if
__name__
==
'__main__'
:
unittest
.
main
()
python/paddle/v2/trainer.py
浏览文件 @
477d92bc
...
...
@@ -174,13 +174,18 @@ class SGD(object):
pass_id
=
pass_id
,
batch_id
=
batch_id
,
cost
=
cost
,
evaluator
=
batch_evaluator
))
evaluator
=
batch_evaluator
,
gm
=
self
.
__gradient_machine__
))
self
.
__parameter_updater__
.
finishBatch
(
cost
)
batch_evaluator
.
finish
()
self
.
__parameter_updater__
.
finishPass
()
pass_evaluator
.
finish
()
event_handler
(
v2_event
.
EndPass
(
pass_id
,
evaluator
=
pass_evaluator
))
event_handler
(
v2_event
.
EndPass
(
pass_id
,
evaluator
=
pass_evaluator
,
gm
=
self
.
__gradient_machine__
))
self
.
__gradient_machine__
.
finish
()
def
test
(
self
,
reader
,
feeding
=
None
):
...
...
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