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
440ad999
P
Paddle
项目概览
PaddlePaddle
/
Paddle
12 个月 前同步成功
通知
2292
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
440ad999
编写于
10月 16, 2017
作者:
T
Tao Luo
提交者:
GitHub
10月 16, 2017
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #4788 from luotao1/seqpool
add SQRT/LAST/FIRST strategy for Seqpool
上级
4da6e86f
6a4282a2
变更
3
隐藏空白更改
内联
并排
Showing
3 changed file
with
97 addition
and
8 deletion
+97
-8
paddle/operators/sequence_pool_op.cc
paddle/operators/sequence_pool_op.cc
+7
-8
paddle/operators/sequence_pool_op.h
paddle/operators/sequence_pool_op.h
+25
-0
python/paddle/v2/framework/tests/test_seq_pool.py
python/paddle/v2/framework/tests/test_seq_pool.py
+65
-0
未找到文件。
paddle/operators/sequence_pool_op.cc
浏览文件 @
440ad999
...
...
@@ -36,11 +36,10 @@ class SequencePoolOpMaker : public framework::OpProtoAndCheckerMaker {
SequencePoolOpMaker
(
framework
::
OpProto
*
proto
,
framework
::
OpAttrChecker
*
op_checker
)
:
OpProtoAndCheckerMaker
(
proto
,
op_checker
)
{
AddInput
(
"X"
,
"A float LoDTensor, the variable-length input of SequencePoolOp"
);
AddOutput
(
"Out"
,
"A float LoDTensor, the variable-length output of SequencePoolOp."
);
AddInput
(
"X"
,
"(LoDTensor), the variable-length input of SequencePoolOp"
);
AddOutput
(
"Out"
,
"(Tensor), output of SequencePoolOp, which does not contain LoD "
"infomation."
);
AddAttr
<
int
>
(
"strategy"
,
"(int, default AVERAGE) the pooling strategy of SequencePoolOp."
)
...
...
@@ -49,13 +48,13 @@ class SequencePoolOpMaker : public framework::OpProtoAndCheckerMaker {
AddComment
(
R"DOC(
SequencePoolOp pools features of all time-steps of each instance.
For a mini-batch of 3 variable
lengths
sentences, containing 2, 3, and 2 time-steps:
For a mini-batch of 3 variable
-length
sentences, containing 2, 3, and 2 time-steps:
Assume X is a [7,M,N]
float LoDTensor, and X->lod()[0] = [0, 2, 5, 7]
.
Assume X is a [7,M,N]
LoDTensor, and X->lod()[0] = [0, 2, 5, 7], 7=2+3+2
.
Besides, for the sake of simplicity, we assume M=1 and N=1,
and the value of X = [[1, 3], [2, 4, 6], [5, 1]].
Thus, Out is a [3,1,1]
float LoDTensor, but Out->lod() is nullptr
.
Thus, Out is a [3,1,1]
Tensor without LoD infomation
.
And for different strategy, the value of Out is as follows:
- AVERAGE: [2, 4, 3], where 2=(1+3)/2, 4=(2+4+6)/3, 3=(5+1)/2
...
...
paddle/operators/sequence_pool_op.h
浏览文件 @
440ad999
...
...
@@ -15,6 +15,7 @@ limitations under the License. */
#pragma once
#include "paddle/framework/eigen.h"
#include "paddle/framework/op_registry.h"
#include "paddle/operators/math/math_function.h"
namespace
paddle
{
namespace
operators
{
...
...
@@ -77,6 +78,16 @@ class SequencePoolKernel : public framework::OpKernel<T> {
case
SUM
:
out_e
.
device
(
place
)
=
in_e
.
sum
(
Eigen
::
array
<
int
,
1
>
({{
0
}}));
break
;
case
SQRT
:
out_e
.
device
(
place
)
=
in_e
.
sum
(
Eigen
::
array
<
int
,
1
>
({{
0
}}))
/
std
::
sqrt
(
static_cast
<
T
>
(
h
));
break
;
case
LAST
:
out_e
.
device
(
place
)
=
in_e
.
chip
(
h
-
1
,
0
);
break
;
case
FIRST
:
out_e
.
device
(
place
)
=
in_e
.
chip
(
0
,
0
);
break
;
default:
PADDLE_THROW
(
"unsupported pooling strategy"
);
}
...
...
@@ -98,6 +109,10 @@ class SequencePoolGradKernel : public framework::OpKernel<T> {
int64_t
w
=
in
->
numel
()
/
dims
[
0
];
in_g
->
mutable_data
<
T
>
(
context
.
GetPlace
());
if
(
strategy
==
LAST
||
strategy
==
FIRST
)
{
// set X@Grad be zero at first when strategy is LAST/FIRST
math
::
SetConstant
<
Place
,
T
>
(
context
.
device_context
(),
in_g
,
0
);
}
auto
place
=
context
.
GetEigenDevice
<
Place
>
();
for
(
int
i
=
0
;
i
<
static_cast
<
int
>
(
lod
.
size
())
-
1
;
++
i
)
{
auto
in_g_t
=
in_g
->
Slice
<
T
>
(
static_cast
<
int
>
(
lod
[
i
]),
...
...
@@ -115,6 +130,16 @@ class SequencePoolGradKernel : public framework::OpKernel<T> {
case
SUM
:
in_g_e
.
device
(
place
)
=
(
out_g_e
).
broadcast
(
bcast
);
break
;
case
SQRT
:
in_g_e
.
device
(
place
)
=
(
out_g_e
/
std
::
sqrt
(
static_cast
<
T
>
(
h
))).
broadcast
(
bcast
);
break
;
case
LAST
:
in_g_e
.
chip
(
h
-
1
,
0
).
device
(
place
)
=
out_g_e
;
break
;
case
FIRST
:
in_g_e
.
chip
(
0
,
0
).
device
(
place
)
=
out_g_e
;
break
;
default:
PADDLE_THROW
(
"unsupported pooling strategy"
);
}
...
...
python/paddle/v2/framework/tests/test_seq_pool.py
浏览文件 @
440ad999
...
...
@@ -82,5 +82,70 @@ class TestSeqSumPool2D(TestSeqAvgPool2D):
out
[
i
]
=
np
.
reshape
(
sub_x
.
sum
(
axis
=
0
),
(
3
,
17
))
class
TestSeqSqrtPool
(
TestSeqAvgPool
):
def
compute
(
self
):
self
.
attrs
=
{
'strategy'
:
SeqPoolType
.
SQRT
}
x
,
lod
=
self
.
inputs
[
'X'
]
out
=
self
.
outputs
[
'Out'
]
for
i
in
range
(
4
):
sub_x
=
x
[
lod
[
0
][
i
]:
lod
[
0
][
i
+
1
],
:]
len
=
lod
[
0
][
i
+
1
]
-
lod
[
0
][
i
]
out
[
i
]
=
sub_x
.
sum
(
axis
=
0
)
/
np
.
sqrt
(
len
)
class
TestSeqSqrtPool2D
(
TestSeqAvgPool2D
):
def
compute
(
self
):
self
.
attrs
=
{
'strategy'
:
SeqPoolType
.
SQRT
}
x
,
lod
=
self
.
inputs
[
'X'
]
out
=
self
.
outputs
[
'Out'
]
for
i
in
range
(
4
):
sub_x
=
np
.
reshape
(
x
[
lod
[
0
][
i
]:
lod
[
0
][
i
+
1
],
:],
(
-
1
,
3
*
17
))
len
=
lod
[
0
][
i
+
1
]
-
lod
[
0
][
i
]
out
[
i
]
=
np
.
reshape
(
sub_x
.
sum
(
axis
=
0
)
/
np
.
sqrt
(
len
),
(
3
,
17
))
def
test_check_grad
(
self
):
self
.
check_grad
([
"X"
],
"Out"
,
max_relative_error
=
0.06
)
class
TestSeqLastPool
(
TestSeqAvgPool
):
def
compute
(
self
):
self
.
attrs
=
{
'strategy'
:
SeqPoolType
.
LAST
}
x
,
lod
=
self
.
inputs
[
'X'
]
out
=
self
.
outputs
[
'Out'
]
for
i
in
range
(
4
):
sub_x
=
x
[
lod
[
0
][
i
]:
lod
[
0
][
i
+
1
],
:]
out
[
i
]
=
sub_x
[
-
1
,
:]
class
TestSeqLastPool2D
(
TestSeqAvgPool2D
):
def
compute
(
self
):
self
.
attrs
=
{
'strategy'
:
SeqPoolType
.
LAST
}
x
,
lod
=
self
.
inputs
[
'X'
]
out
=
self
.
outputs
[
'Out'
]
for
i
in
range
(
4
):
sub_x
=
np
.
reshape
(
x
[
lod
[
0
][
i
]:
lod
[
0
][
i
+
1
],
:],
(
-
1
,
3
*
17
))
out
[
i
]
=
np
.
reshape
(
sub_x
[
-
1
,
:],
(
3
,
17
))
class
TestSeqFirstPool
(
TestSeqAvgPool
):
def
compute
(
self
):
self
.
attrs
=
{
'strategy'
:
SeqPoolType
.
FIRST
}
x
,
lod
=
self
.
inputs
[
'X'
]
out
=
self
.
outputs
[
'Out'
]
for
i
in
range
(
4
):
sub_x
=
x
[
lod
[
0
][
i
]:
lod
[
0
][
i
+
1
],
:]
out
[
i
]
=
sub_x
[
0
,
:]
class
TestSeqFirstPool2D
(
TestSeqAvgPool2D
):
def
compute
(
self
):
self
.
attrs
=
{
'strategy'
:
SeqPoolType
.
FIRST
}
x
,
lod
=
self
.
inputs
[
'X'
]
out
=
self
.
outputs
[
'Out'
]
for
i
in
range
(
4
):
sub_x
=
np
.
reshape
(
x
[
lod
[
0
][
i
]:
lod
[
0
][
i
+
1
],
:],
(
-
1
,
3
*
17
))
out
[
i
]
=
np
.
reshape
(
sub_x
[
0
,
:],
(
3
,
17
))
if
__name__
==
'__main__'
:
unittest
.
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