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
Paddle
提交
a8c6ce9b
P
Paddle
项目概览
PaddlePaddle
/
Paddle
大约 1 年 前同步成功
通知
2298
Star
20931
Fork
5422
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1423
列表
看板
标记
里程碑
合并请求
543
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
1,423
Issue
1,423
列表
看板
标记
里程碑
合并请求
543
合并请求
543
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
a8c6ce9b
编写于
9月 28, 2017
作者:
Y
Yu Yang
浏览文件
操作
浏览文件
下载
差异文件
Merge branch 'develop' of github.com:baidu/Paddle into feature/BetterActivationKern
上级
337b7ebe
9fbf94b6
变更
15
隐藏空白更改
内联
并排
Showing
15 changed file
with
290 addition
and
73 deletion
+290
-73
paddle/framework/CMakeLists.txt
paddle/framework/CMakeLists.txt
+1
-1
paddle/framework/grad_op_builder.cc
paddle/framework/grad_op_builder.cc
+39
-0
paddle/framework/grad_op_builder.h
paddle/framework/grad_op_builder.h
+3
-0
paddle/framework/grad_op_builder_test.cc
paddle/framework/grad_op_builder_test.cc
+79
-0
paddle/framework/op_desc.cc
paddle/framework/op_desc.cc
+11
-0
paddle/framework/op_desc.h
paddle/framework/op_desc.h
+6
-0
paddle/operators/activation_op.cc
paddle/operators/activation_op.cc
+14
-0
paddle/operators/activation_op.h
paddle/operators/activation_op.h
+22
-1
paddle/pybind/pybind.cc
paddle/pybind/pybind.cc
+7
-9
paddle/pybind/tensor_py.h
paddle/pybind/tensor_py.h
+14
-1
python/paddle/v2/framework/tests/op_test.py
python/paddle/v2/framework/tests/op_test.py
+60
-42
python/paddle/v2/framework/tests/test_activation_op.py
python/paddle/v2/framework/tests/test_activation_op.py
+17
-0
python/paddle/v2/framework/tests/test_cross_entropy_op.py
python/paddle/v2/framework/tests/test_cross_entropy_op.py
+1
-1
python/paddle/v2/framework/tests/test_elementwise_mul_op.py
python/paddle/v2/framework/tests/test_elementwise_mul_op.py
+15
-17
python/paddle/v2/framework/tests/test_prelu_op.py
python/paddle/v2/framework/tests/test_prelu_op.py
+1
-1
未找到文件。
paddle/framework/CMakeLists.txt
浏览文件 @
a8c6ce9b
...
...
@@ -26,7 +26,7 @@ cc_library(op_info SRCS op_info.cc DEPS attribute framework_proto)
cc_library
(
operator SRCS operator.cc DEPS op_info device_context tensor scope
)
cc_test
(
operator_test SRCS operator_test.cc DEPS operator op_registry
)
cc_library
(
grad_op_builder SRCS grad_op_builder.cc DEPS operator
)
cc_library
(
grad_op_builder SRCS grad_op_builder.cc DEPS operator
proto_desc
)
cc_library
(
op_registry SRCS op_registry.cc DEPS grad_op_builder op_proto_maker op_info
)
cc_test
(
op_registry_test SRCS op_registry_test.cc DEPS op_registry
)
cc_test
(
grad_op_builder_test SRCS grad_op_builder_test.cc DEPS grad_op_builder op_registry add_op
)
...
...
paddle/framework/grad_op_builder.cc
浏览文件 @
a8c6ce9b
...
...
@@ -54,5 +54,44 @@ OperatorBase* BuildGradOp(const OperatorBase* op) {
return
grad_info
.
Creator
()(
info
.
grad_op_type_
,
inputs
,
outputs
,
op
->
Attrs
());
}
static
void
TransOpDescArg
(
const
OpDescBind
*
src_op
,
const
OpArgType
&
src_type
,
bool
is_grad
,
OpDescBind
*
dst_op
,
const
OpArgType
&
dst_type
)
{
PADDLE_ENFORCE
(
dst_op
!=
nullptr
,
"Protobuf desc of gradient op must be initialized first."
);
const
auto
&
proto
=
OpInfoMap
::
Instance
().
Get
(
src_op
->
Type
()).
Proto
();
const
auto
&
src_arg_list
=
src_type
==
OpArgType
::
IN
?
proto
.
inputs
()
:
proto
.
outputs
();
for
(
const
auto
&
arg
:
src_arg_list
)
{
if
(
arg
.
not_in_gradient
()
&&
!
is_grad
)
continue
;
const
std
::
string
src_name
=
arg
.
name
();
std
::
vector
<
std
::
string
>
vars
=
src_type
==
OpArgType
::
IN
?
src_op
->
Input
(
src_name
)
:
src_op
->
Output
(
src_name
);
if
(
is_grad
)
{
for
(
std
::
string
&
var
:
vars
)
{
var
=
GradVarName
(
var
);
}
}
std
::
string
dst_name
=
is_grad
?
GradVarName
(
src_name
)
:
src_name
;
dst_type
==
OpArgType
::
IN
?
dst_op
->
SetInput
(
dst_name
,
vars
)
:
dst_op
->
SetOutput
(
dst_name
,
vars
);
}
}
void
CompleteGradOpDesc
(
const
OpDescBind
*
forw_op
,
OpDescBind
*
grad_op
)
{
auto
&
info
=
OpInfoMap
::
Instance
().
Get
(
forw_op
->
Type
());
PADDLE_ENFORCE
(
info
.
HasGradientOp
());
grad_op
->
SetType
(
info
.
grad_op_type_
);
TransOpDescArg
(
forw_op
,
OpArgType
::
IN
,
false
,
grad_op
,
OpArgType
::
IN
);
TransOpDescArg
(
forw_op
,
OpArgType
::
OUT
,
false
,
grad_op
,
OpArgType
::
IN
);
TransOpDescArg
(
forw_op
,
OpArgType
::
OUT
,
true
,
grad_op
,
OpArgType
::
IN
);
TransOpDescArg
(
forw_op
,
OpArgType
::
IN
,
true
,
grad_op
,
OpArgType
::
OUT
);
grad_op
->
SetAttrMap
(
forw_op
->
GetAttrMap
());
}
}
// namespace framework
}
// namespace paddle
paddle/framework/grad_op_builder.h
浏览文件 @
a8c6ce9b
...
...
@@ -14,6 +14,7 @@ limitations under the License. */
#pragma once
#include "paddle/framework/op_desc.h"
#include "paddle/framework/operator.h"
namespace
paddle
{
...
...
@@ -21,5 +22,7 @@ namespace framework {
OperatorBase
*
BuildGradOp
(
const
OperatorBase
*
op
);
void
CompleteGradOpDesc
(
const
OpDescBind
*
forw_op
,
OpDescBind
*
grad_op
);
}
// namespace framework
}
// namespace paddle
paddle/framework/grad_op_builder_test.cc
浏览文件 @
a8c6ce9b
...
...
@@ -120,3 +120,82 @@ TEST(GradOpBuilder, IOIgnoredInGradient) {
std
::
vector
<
std
::
string
>
(
{
f
::
GradVarName
(
"in3_1"
),
f
::
GradVarName
(
"in3_2"
)}));
}
TEST
(
GradOpDescBuilder
,
MutiInOut
)
{
f
::
OpDescBind
*
forw_op
=
new
f
::
OpDescBind
();
forw_op
->
SetType
(
"mult_io"
);
forw_op
->
SetInput
(
"In1"
,
{
"in1"
});
forw_op
->
SetInput
(
"In2_mult"
,
{
"in2_1"
,
"in2_2"
,
"in2_3"
});
forw_op
->
SetInput
(
"In3"
,
{
"in3"
});
forw_op
->
SetOutput
(
"Out1"
,
{
"out1"
});
forw_op
->
SetOutput
(
"Out2_mult"
,
{
"out2_1"
,
"out2_2"
});
f
::
OpDescBind
*
grad_op
=
new
f
::
OpDescBind
();
f
::
CompleteGradOpDesc
(
forw_op
,
grad_op
);
EXPECT_EQ
(
grad_op
->
Type
(),
"mult_io_grad"
);
ASSERT_EQ
(
grad_op
->
InputNames
().
size
(),
3UL
+
2UL
+
2UL
);
EXPECT_EQ
(
grad_op
->
Input
(
"In1"
),
std
::
vector
<
std
::
string
>
({
"in1"
}));
EXPECT_EQ
(
grad_op
->
Input
(
"In2_mult"
),
std
::
vector
<
std
::
string
>
({
"in2_1"
,
"in2_2"
,
"in2_3"
}));
EXPECT_EQ
(
grad_op
->
Input
(
"In3"
),
std
::
vector
<
std
::
string
>
({
"in3"
}));
EXPECT_EQ
(
grad_op
->
Input
(
"Out1"
),
std
::
vector
<
std
::
string
>
({
"out1"
}));
EXPECT_EQ
(
grad_op
->
Input
(
"Out2_mult"
),
std
::
vector
<
std
::
string
>
({
"out2_1"
,
"out2_2"
}));
EXPECT_EQ
(
grad_op
->
Input
(
f
::
GradVarName
(
"Out1"
)),
std
::
vector
<
std
::
string
>
({
f
::
GradVarName
(
"out1"
)}));
EXPECT_EQ
(
grad_op
->
Input
(
f
::
GradVarName
(
"Out2_mult"
)),
std
::
vector
<
std
::
string
>
(
{
f
::
GradVarName
(
"out2_1"
),
f
::
GradVarName
(
"out2_2"
)}));
ASSERT_EQ
(
grad_op
->
OutputNames
().
size
(),
3UL
);
EXPECT_EQ
(
grad_op
->
Output
(
f
::
GradVarName
(
"In1"
)),
std
::
vector
<
std
::
string
>
({
f
::
GradVarName
(
"in1"
)}));
EXPECT_EQ
(
grad_op
->
Output
(
f
::
GradVarName
(
"In2_mult"
)),
std
::
vector
<
std
::
string
>
({
f
::
GradVarName
(
"in2_1"
),
f
::
GradVarName
(
"in2_2"
),
f
::
GradVarName
(
"in2_3"
)}));
EXPECT_EQ
(
grad_op
->
Output
(
f
::
GradVarName
(
"In3"
)),
std
::
vector
<
std
::
string
>
({
f
::
GradVarName
(
"in3"
)}));
delete
forw_op
;
delete
grad_op
;
}
TEST
(
GradOpDescBuilder
,
IOIgnoredInGradient
)
{
f
::
OpDescBind
*
forw_op
=
new
f
::
OpDescBind
();
forw_op
->
SetType
(
"io_ignored"
);
forw_op
->
SetInput
(
"In1"
,
{
"in1"
});
forw_op
->
SetInput
(
"In2_mult"
,
{
"in2_1"
,
"in2_2"
});
forw_op
->
SetInput
(
"In3_mult"
,
{
"in3_1"
,
"in3_2"
});
forw_op
->
SetOutput
(
"Out1_mult"
,
{
"out1_1"
,
"out1_2"
});
forw_op
->
SetOutput
(
"Out2"
,
{
"out2"
});
f
::
OpDescBind
*
grad_op
=
new
f
::
OpDescBind
();
f
::
CompleteGradOpDesc
(
forw_op
,
grad_op
);
EXPECT_EQ
(
grad_op
->
Type
(),
"io_ignored_grad"
);
// 'In2' and 'Out2' are ignored in gradient calculating
ASSERT_EQ
(
grad_op
->
InputNames
().
size
(),
2UL
+
1UL
+
2UL
);
EXPECT_EQ
(
grad_op
->
Input
(
"In1"
),
std
::
vector
<
std
::
string
>
({
"in1"
}));
EXPECT_EQ
(
grad_op
->
Input
(
"In3_mult"
),
std
::
vector
<
std
::
string
>
({
"in3_1"
,
"in3_2"
}));
EXPECT_EQ
(
grad_op
->
Input
(
"Out1_mult"
),
std
::
vector
<
std
::
string
>
({
"out1_1"
,
"out1_2"
}));
EXPECT_EQ
(
grad_op
->
Input
(
f
::
GradVarName
(
"Out1_mult"
)),
std
::
vector
<
std
::
string
>
(
{
f
::
GradVarName
(
"out1_1"
),
f
::
GradVarName
(
"out1_2"
)}));
EXPECT_EQ
(
grad_op
->
Input
(
f
::
GradVarName
(
"Out2"
)),
std
::
vector
<
std
::
string
>
({
f
::
GradVarName
(
"out2"
)}));
ASSERT_EQ
(
grad_op
->
OutputNames
().
size
(),
3UL
);
EXPECT_EQ
(
grad_op
->
Output
(
f
::
GradVarName
(
"In1"
)),
std
::
vector
<
std
::
string
>
({
f
::
GradVarName
(
"in1"
)}));
EXPECT_EQ
(
grad_op
->
Output
(
f
::
GradVarName
(
"In2_mult"
)),
std
::
vector
<
std
::
string
>
(
{
f
::
GradVarName
(
"in2_1"
),
f
::
GradVarName
(
"in2_2"
)}));
EXPECT_EQ
(
grad_op
->
Output
(
f
::
GradVarName
(
"In3_mult"
)),
std
::
vector
<
std
::
string
>
(
{
f
::
GradVarName
(
"in3_1"
),
f
::
GradVarName
(
"in3_2"
)}));
delete
forw_op
;
delete
grad_op
;
}
\ No newline at end of file
paddle/framework/op_desc.cc
浏览文件 @
a8c6ce9b
...
...
@@ -89,6 +89,12 @@ void OpDescBind::SetAttr(const std::string &name, const Attribute &v) {
need_update_
=
true
;
}
void
OpDescBind
::
SetAttrMap
(
const
std
::
unordered_map
<
std
::
string
,
Attribute
>
&
attr_map
)
{
attrs_
=
attr_map
;
need_update_
=
true
;
}
Attribute
OpDescBind
::
GetAttr
(
const
std
::
string
&
name
)
const
{
auto
it
=
attrs_
.
find
(
name
);
PADDLE_ENFORCE
(
it
!=
attrs_
.
end
(),
"Attribute %s is not found"
,
name
);
...
...
@@ -101,6 +107,11 @@ int OpDescBind::GetBlockAttr(const std::string &name) const {
return
boost
::
get
<
BlockDesc
*>
(
it
->
second
)
->
idx
();
}
const
std
::
unordered_map
<
std
::
string
,
Attribute
>
&
OpDescBind
::
GetAttrMap
()
const
{
return
attrs_
;
}
void
OpDescBind
::
Sync
()
{
if
(
need_update_
)
{
this
->
op_desc_
.
mutable_inputs
()
->
Clear
();
...
...
paddle/framework/op_desc.h
浏览文件 @
a8c6ce9b
...
...
@@ -60,10 +60,16 @@ class OpDescBind {
void
SetBlockAttr
(
const
std
::
string
&
name
,
BlockDescBind
&
block
);
// Only be used in C++
void
SetAttrMap
(
const
std
::
unordered_map
<
std
::
string
,
Attribute
>
&
attr_map
);
Attribute
GetAttr
(
const
std
::
string
&
name
)
const
;
int
GetBlockAttr
(
const
std
::
string
&
name
)
const
;
// Only be used in C++
const
std
::
unordered_map
<
std
::
string
,
Attribute
>
&
GetAttrMap
()
const
;
private:
struct
SetAttrDescVisitor
:
public
boost
::
static_visitor
<
void
>
{
explicit
SetAttrDescVisitor
(
OpDesc
::
Attr
*
attr
)
:
attr_
(
attr
)
{}
...
...
paddle/operators/activation_op.cc
浏览文件 @
a8c6ce9b
...
...
@@ -132,6 +132,17 @@ class SquareOpMaker : public framework::OpProtoAndCheckerMaker {
}
};
class
SoftsignOpMaker
:
public
framework
::
OpProtoAndCheckerMaker
{
public:
SoftsignOpMaker
(
framework
::
OpProto
*
proto
,
framework
::
OpAttrChecker
*
op_checker
)
:
OpProtoAndCheckerMaker
(
proto
,
op_checker
)
{
AddInput
(
"X"
,
"Input of Softsign operator"
);
AddOutput
(
"Y"
,
"Output of Softsign operator"
);
AddComment
(
"Softsign activation operator, softsign(x) = x / (1 + |x|)"
);
}
};
template
<
typename
AttrType
>
class
BReluOpMaker
:
public
framework
::
OpProtoAndCheckerMaker
{
public:
...
...
@@ -223,6 +234,9 @@ REGISTER_OP(log, ops::ActivationOp, ops::LogOpMaker, log_grad,
REGISTER_OP
(
square
,
ops
::
ActivationOp
,
ops
::
SquareOpMaker
,
square_grad
,
ops
::
ActivationOpGrad
);
REGISTER_OP
(
softsign
,
ops
::
ActivationOp
,
ops
::
SoftsignOpMaker
,
softsign_grad
,
ops
::
ActivationOpGrad
);
REGISTER_OP
(
brelu
,
ops
::
ActivationOp
,
ops
::
BReluOpMaker
<
float
>
,
brelu_grad
,
ops
::
ActivationOpGrad
);
...
...
paddle/operators/activation_op.h
浏览文件 @
a8c6ce9b
...
...
@@ -262,6 +262,26 @@ struct BReluGradFunctor : public BaseActivationFunctor<T> {
}
};
// softsign(x) = x / (1 + |x|)
template
<
typename
T
>
struct
SoftsignFunctor
:
public
BaseActivationFunctor
<
T
>
{
template
<
typename
Device
,
typename
X
,
typename
Y
>
void
operator
()(
Device
d
,
X
x
,
Y
y
)
{
y
.
device
(
d
)
=
x
/
(
static_cast
<
T
>
(
1
)
+
x
.
abs
());
}
};
// d(softsign(x))/dx = 1 / (1 + |x|)^2
// Taken from https://en.wikipedia.org/wiki/Activation_function
template
<
typename
T
>
struct
SoftsignGradFunctor
:
public
BaseActivationFunctor
<
T
>
{
template
<
typename
Device
,
typename
X
,
typename
Y
,
typename
dY
,
typename
dX
>
void
operator
()(
Device
d
,
X
x
,
Y
y
,
dY
dy
,
dX
dx
)
{
dx
.
device
(
d
)
=
dy
*
(
static_cast
<
T
>
(
1
)
/
(
static_cast
<
T
>
(
1
)
+
x
.
abs
()).
square
());
}
};
template
<
typename
T
>
struct
SoftReluFunctor
:
public
BaseActivationFunctor
<
T
>
{
float
threshold
;
...
...
@@ -358,4 +378,5 @@ struct STanhGradFunctor : public BaseActivationFunctor<T> {
__macro(brelu, BReluFunctor, BReluGradFunctor); \
__macro(soft_relu, SoftReluFunctor, SoftReluGradFunctor); \
__macro(pow, PowFunctor, PowGradFunctor); \
__macro(stanh, STanhFunctor, STanhGradFunctor)
__macro(stanh, STanhFunctor, STanhGradFunctor); \
__macro(softsign, SoftsignFunctor, SoftsignGradFunctor)
paddle/pybind/pybind.cc
浏览文件 @
a8c6ce9b
...
...
@@ -77,20 +77,18 @@ PYBIND11_PLUGIN(core) {
})
.
def
(
"set"
,
PyCPUTensorSetFromArray
<
float
>
)
.
def
(
"set"
,
PyCPUTensorSetFromArray
<
int
>
)
.
def
(
"set"
,
PyCPUTensorSetFromArray
<
double
>
)
#ifndef PADDLE_ONLY_CPU
.
def
(
"set"
,
PyCUDATensorSetFromArray
<
float
>
)
.
def
(
"set"
,
PyCUDATensorSetFromArray
<
int
>
)
.
def
(
"set"
,
PyCUDATensorSetFromArray
<
double
>
)
#endif
.
def
(
"shape"
,
[](
Tensor
&
self
)
{
return
vectorize
(
self
.
dims
());
})
.
def
(
"set_float_element"
,
[](
Tensor
&
self
,
size_t
offset
,
float
f
)
{
// TODO(yuyang18): Only support GPU now.
self
.
data
<
float
>
()[
offset
]
=
f
;
})
.
def
(
"get_float_element"
,
[](
Tensor
&
self
,
size_t
offset
)
->
float
{
// TODO(yuyang18): Only support GPU now.
return
self
.
data
<
float
>
()[
offset
];
});
.
def
(
"set_float_element"
,
TensorSetElement
<
float
>
)
.
def
(
"get_float_element"
,
TensorGetElement
<
float
>
)
.
def
(
"set_double_element"
,
TensorSetElement
<
double
>
)
.
def
(
"get_double_element"
,
TensorGetElement
<
double
>
)
.
def
(
"dtype"
,
[](
Tensor
&
self
)
{
return
ToDataType
(
self
.
type
());
});
py
::
class_
<
LoDTensor
,
Tensor
>
(
m
,
"LoDTensor"
)
.
def_buffer
(
...
...
paddle/pybind/tensor_py.h
浏览文件 @
a8c6ce9b
...
...
@@ -73,10 +73,23 @@ struct CastToPyBufferImpl<true, I, ARGS...> {
};
}
// namespace details
inline
py
::
buffer_info
CastToPyBuffer
(
framework
::
Tensor
&
tensor
)
{
auto
buffer_info
=
details
::
CastToPyBufferImpl
<
true
,
0
,
float
,
int
>
()(
tensor
);
auto
buffer_info
=
details
::
CastToPyBufferImpl
<
true
,
0
,
float
,
int
,
double
>
()(
tensor
);
return
buffer_info
;
}
template
<
typename
T
>
T
TensorGetElement
(
framework
::
Tensor
&
self
,
size_t
offset
)
{
PADDLE_ENFORCE
(
platform
::
is_cpu_place
(
self
.
place
()));
return
self
.
data
<
T
>
()[
offset
];
}
template
<
typename
T
>
void
TensorSetElement
(
framework
::
Tensor
&
self
,
size_t
offset
,
T
elem
)
{
PADDLE_ENFORCE
(
platform
::
is_cpu_place
(
self
.
place
()));
self
.
data
<
T
>
()[
offset
]
=
elem
;
}
template
<
typename
T
>
void
PyCPUTensorSetFromArray
(
framework
::
Tensor
&
self
,
...
...
python/paddle/v2/framework/tests/op_test.py
浏览文件 @
a8c6ce9b
...
...
@@ -12,17 +12,19 @@ def grad_var_name(var_name):
def
create_op
(
scope
,
op_type
,
inputs
,
outputs
,
attrs
):
kwargs
=
dict
()
def
__create_var__
(
name
,
var_name
):
scope
.
new_var
(
var_name
)
kwargs
[
name
].
append
(
var_name
)
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
)
__create_var__
(
in_name
,
sub_in_name
)
else
:
var
=
scope
.
new_var
(
in_name
)
kwargs
[
in_name
].
append
(
in_name
)
__create_var__
(
in_name
,
in_name
)
for
out_name
,
out_dup
in
Operator
.
get_op_outputs
(
op_type
):
if
out_name
in
outputs
:
...
...
@@ -30,11 +32,9 @@ def create_op(scope, op_type, inputs, outputs, attrs):
if
out_dup
:
sub_out
=
outputs
[
out_name
]
for
sub_out_name
,
_
in
sub_out
:
var
=
scope
.
new_var
(
sub_out_name
)
kwargs
[
out_name
].
append
(
sub_out_name
)
__create_var__
(
out_name
,
sub_out_name
)
else
:
var
=
scope
.
new_var
(
out_name
)
kwargs
[
out_name
].
append
(
out_name
)
__create_var__
(
out_name
,
out_name
)
for
attr_name
in
Operator
.
get_op_attr_names
(
op_type
):
if
attr_name
in
attrs
:
...
...
@@ -44,49 +44,46 @@ def create_op(scope, op_type, inputs, outputs, attrs):
def
set_input
(
scope
,
op
,
inputs
,
place
):
def
__set_input__
(
var_name
,
var
):
tensor
=
scope
.
find_var
(
var_name
).
get_tensor
()
if
isinstance
(
var
,
tuple
):
tensor
.
set_lod
(
var
[
1
])
var
=
var
[
0
]
tensor
.
set_dims
(
var
.
shape
)
tensor
.
set
(
var
,
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
,
sub_in_val
in
sub_in
:
var
=
scope
.
find_var
(
sub_in_name
)
tensor
=
var
.
get_tensor
()
sub_in_array
=
sub_in_val
[
0
]
\
if
isinstance
(
sub_in_val
,
tuple
)
else
sub_in_val
tensor
.
set_dims
(
sub_in_array
.
shape
)
tensor
.
set
(
sub_in_array
,
place
)
if
isinstance
(
sub_in_val
,
tuple
):
tensor
.
set_lod
(
sub_in_val
[
1
])
__set_input__
(
sub_in_name
,
sub_in_val
)
else
:
var
=
scope
.
find_var
(
in_name
)
tensor
=
var
.
get_tensor
()
in_val
=
inputs
[
in_name
]
in_array
=
in_val
[
0
]
if
isinstance
(
in_val
,
tuple
)
else
in_val
tensor
.
set_dims
(
in_array
.
shape
)
tensor
.
set
(
in_array
,
place
)
if
isinstance
(
in_val
,
tuple
):
tensor
.
set_lod
(
in_val
[
1
])
__set_input__
(
in_name
,
inputs
[
in_name
])
def
set_output_grad
(
scope
,
op
,
outputs
,
place
):
def
__set_tensor__
(
name
):
out_tensor
=
scope
.
find_var
(
name
).
get_tensor
()
grad_tensor
=
scope
.
new_var
(
grad_var_name
(
name
)).
get_tensor
()
out_dtype
=
out_tensor
.
dtype
()
if
out_dtype
==
core
.
DataType
.
FP64
:
data
=
np
.
ones
(
out_tensor
.
shape
(),
dtype
=
np
.
float64
)
elif
out_dtype
==
core
.
DataType
.
FP32
:
data
=
np
.
ones
(
out_tensor
.
shape
(),
dtype
=
np
.
float32
)
else
:
raise
ValueError
(
"Not supported data type "
+
str
(
out_dtype
))
grad_tensor
.
set
(
data
,
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
)
__set_tensor__
(
sub_out_name
)
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
)
__set_tensor__
(
out_name
)
def
get_numeric_gradient
(
scope
,
...
...
@@ -96,7 +93,6 @@ def get_numeric_gradient(scope,
output_names
,
delta
=
0.005
,
in_place
=
False
):
set_input
(
scope
,
op
,
inputs
,
core
.
CPUPlace
())
tensor_to_check
=
scope
.
find_var
(
input_to_check
).
get_tensor
()
...
...
@@ -115,7 +111,29 @@ def get_numeric_gradient(scope,
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'
)
tensor_to_check_dtype
=
tensor_to_check
.
dtype
()
if
tensor_to_check_dtype
==
core
.
DataType
.
FP32
:
tensor_to_check_dtype
=
np
.
float32
elif
tensor_to_check_dtype
==
core
.
DataType
.
FP64
:
tensor_to_check_dtype
=
np
.
float64
else
:
raise
ValueError
(
"Not supported data type "
+
str
(
tensor_to_check_dtype
))
gradient_flat
=
np
.
zeros
(
shape
=
(
tensor_size
,
),
dtype
=
tensor_to_check_dtype
)
def
__get_elem__
(
tensor
,
i
):
if
tensor_to_check_dtype
==
np
.
float32
:
return
tensor
.
get_float_element
(
i
)
else
:
return
tensor
.
get_double_element
(
i
)
def
__set_elem__
(
tensor
,
i
,
e
):
if
tensor_to_check_dtype
==
np
.
float32
:
tensor
.
set_float_element
(
i
,
e
)
else
:
tensor
.
set_double_element
(
i
,
e
)
# 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
):
...
...
@@ -123,20 +141,20 @@ def get_numeric_gradient(scope,
set_input
(
scope
,
op
,
inputs
,
core
.
CPUPlace
())
# get one input element throw it's index i.
origin
=
tensor_to_check
.
get_float_element
(
i
)
origin
=
__get_elem__
(
tensor_to_check
,
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
)
__set_elem__
(
tensor_to_check
,
i
,
x_pos
)
y_pos
=
get_output
()
if
in_place
:
set_input
(
scope
,
op
,
inputs
,
core
.
CPUPlace
())
x_neg
=
origin
-
delta
tensor_to_check
.
set_float_element
(
i
,
x_neg
)
__set_elem__
(
tensor_to_check
,
i
,
x_neg
)
y_neg
=
get_output
()
tensor_to_check
.
set_float_element
(
i
,
origin
)
__set_elem__
(
tensor_to_check
,
i
,
origin
)
gradient_flat
[
i
]
=
(
y_pos
-
y_neg
)
/
delta
/
2
return
gradient_flat
.
reshape
(
tensor_to_check
.
get_dims
())
...
...
python/paddle/v2/framework/tests/test_activation_op.py
浏览文件 @
a8c6ce9b
...
...
@@ -219,5 +219,22 @@ class TestSTanh(OpTest):
self
.
check_grad
([
'X'
],
'Y'
,
max_relative_error
=
0.007
)
class
TestSoftsign
(
OpTest
):
def
setUp
(
self
):
self
.
op_type
=
"softsign"
self
.
inputs
=
{
'X'
:
np
.
random
.
uniform
(
-
1
,
1
,
[
11
,
17
]).
astype
(
"float32"
)
}
self
.
outputs
=
{
'Y'
:
np
.
divide
(
self
.
inputs
[
'X'
],
1
+
np
.
abs
(
self
.
inputs
[
'X'
]))
}
def
test_check_output
(
self
):
self
.
check_output
()
def
test_check_grad
(
self
):
self
.
check_grad
([
'X'
],
'Y'
,
max_relative_error
=
0.007
)
if
__name__
==
"__main__"
:
unittest
.
main
()
python/paddle/v2/framework/tests/test_cross_entropy_op.py
浏览文件 @
a8c6ce9b
...
...
@@ -80,7 +80,7 @@ class TestCrossEntropyOp3(OpTest):
cross_entropy2
=
(
-
label
*
np
.
log
(
X
)).
sum
(
axis
=
1
,
keepdims
=
True
).
astype
(
"float32"
)
self
.
inputs
=
{
"X"
:
X
,
"Label"
:
label
}
self
.
inputs
=
{
"X"
:
X
,
"Label"
:
label
.
astype
(
np
.
float32
)
}
self
.
outputs
=
{
"Y"
:
cross_entropy
}
self
.
attrs
=
{
"softLabel"
:
True
}
...
...
python/paddle/v2/framework/tests/test_elementwise_mul_op.py
浏览文件 @
a8c6ce9b
...
...
@@ -7,8 +7,8 @@ class ElementwiseMulOp(OpTest):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_mul"
self
.
inputs
=
{
'X'
:
np
.
random
.
uniform
(
0.1
,
1
,
[
13
,
17
]).
astype
(
"float
32
"
),
'Y'
:
np
.
random
.
uniform
(
0.1
,
1
,
[
13
,
17
]).
astype
(
"float
32
"
)
'X'
:
np
.
random
.
uniform
(
0.1
,
1
,
[
13
,
17
]).
astype
(
"float
64
"
),
'Y'
:
np
.
random
.
uniform
(
0.1
,
1
,
[
13
,
17
]).
astype
(
"float
64
"
)
}
self
.
outputs
=
{
'Out'
:
np
.
multiply
(
self
.
inputs
[
'X'
],
self
.
inputs
[
'Y'
])}
...
...
@@ -16,23 +16,21 @@ class ElementwiseMulOp(OpTest):
self
.
check_output
()
def
test_check_grad_normal
(
self
):
self
.
check_grad
([
'X'
,
'Y'
],
'Out'
,
max_relative_error
=
0.1
)
self
.
check_grad
([
'X'
,
'Y'
],
'Out'
)
def
test_check_grad_ingore_x
(
self
):
self
.
check_grad
(
[
'Y'
],
'Out'
,
max_relative_error
=
0.1
,
no_grad_set
=
set
(
"X"
))
self
.
check_grad
([
'Y'
],
'Out'
,
no_grad_set
=
set
(
"X"
))
def
test_check_grad_ingore_y
(
self
):
self
.
check_grad
(
[
'X'
],
'Out'
,
max_relative_error
=
0.1
,
no_grad_set
=
set
(
'Y'
))
self
.
check_grad
([
'X'
],
'Out'
,
no_grad_set
=
set
(
'Y'
))
class
TestElementwiseMulOp_Vector
(
ElementwiseMulOp
):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_mul"
self
.
inputs
=
{
'X'
:
np
.
random
.
random
((
32
,
)).
astype
(
"float
32
"
),
'Y'
:
np
.
random
.
random
((
32
,
)).
astype
(
"float
32
"
)
'X'
:
np
.
random
.
random
((
32
,
)).
astype
(
"float
64
"
),
'Y'
:
np
.
random
.
random
((
32
,
)).
astype
(
"float
64
"
)
}
self
.
outputs
=
{
'Out'
:
np
.
multiply
(
self
.
inputs
[
'X'
],
self
.
inputs
[
'Y'
])}
...
...
@@ -41,8 +39,8 @@ class TestElementwiseMulOp_broadcast_0(ElementwiseMulOp):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_mul"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
np
.
float
32
),
'Y'
:
np
.
random
.
rand
(
2
).
astype
(
np
.
float
32
)
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
np
.
float
64
),
'Y'
:
np
.
random
.
rand
(
2
).
astype
(
np
.
float
64
)
}
self
.
attrs
=
{
'axis'
:
0
}
...
...
@@ -55,8 +53,8 @@ class TestElementwiseMulOp_broadcast_1(ElementwiseMulOp):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_mul"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
np
.
float
32
),
'Y'
:
np
.
random
.
rand
(
3
).
astype
(
np
.
float
32
)
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
np
.
float
64
),
'Y'
:
np
.
random
.
rand
(
3
).
astype
(
np
.
float
64
)
}
self
.
attrs
=
{
'axis'
:
1
}
...
...
@@ -69,8 +67,8 @@ class TestElementwiseMulOp_broadcast_2(ElementwiseMulOp):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_mul"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
np
.
float
32
),
'Y'
:
np
.
random
.
rand
(
4
).
astype
(
np
.
float
32
)
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
).
astype
(
np
.
float
64
),
'Y'
:
np
.
random
.
rand
(
4
).
astype
(
np
.
float
64
)
}
self
.
outputs
=
{
...
...
@@ -82,8 +80,8 @@ class TestElementwiseMulOp_broadcast_3(ElementwiseMulOp):
def
setUp
(
self
):
self
.
op_type
=
"elementwise_mul"
self
.
inputs
=
{
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
,
5
).
astype
(
np
.
float
32
),
'Y'
:
np
.
random
.
rand
(
3
,
4
).
astype
(
np
.
float
32
)
'X'
:
np
.
random
.
rand
(
2
,
3
,
4
,
5
).
astype
(
np
.
float
64
),
'Y'
:
np
.
random
.
rand
(
3
,
4
).
astype
(
np
.
float
64
)
}
self
.
attrs
=
{
'axis'
:
1
}
...
...
python/paddle/v2/framework/tests/test_prelu_op.py
浏览文件 @
a8c6ce9b
...
...
@@ -17,7 +17,7 @@ class PReluTest(OpTest):
x_np_sign
=
np
.
sign
(
x_np
)
x_np
=
x_np_sign
*
np
.
maximum
(
x_np
,
.
005
)
alpha_np
=
np
.
array
([.
1
])
alpha_np
=
np
.
array
([.
1
]
,
dtype
=
"float32"
)
self
.
inputs
=
{
'X'
:
x_np
,
'Alpha'
:
alpha_np
}
out_np
=
np
.
maximum
(
self
.
inputs
[
'X'
],
0.
)
out_np
=
out_np
+
np
.
minimum
(
self
.
inputs
[
'X'
],
...
...
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