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
PaddleDetection
提交
12f2b8eb
P
PaddleDetection
项目概览
PaddlePaddle
/
PaddleDetection
12 个月 前同步成功
通知
692
Star
11112
Fork
2696
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
184
列表
看板
标记
里程碑
合并请求
40
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
PaddleDetection
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
184
Issue
184
列表
看板
标记
里程碑
合并请求
40
合并请求
40
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
12f2b8eb
编写于
9月 25, 2017
作者:
L
Liu Yiqun
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Correct the forward of sequence_softmax_op.
上级
4d929394
变更
3
隐藏空白更改
内联
并排
Showing
3 changed file
with
12 addition
and
7 deletion
+12
-7
paddle/operators/reshape_op.cc
paddle/operators/reshape_op.cc
+1
-2
paddle/operators/sequence_softmax_op.cc
paddle/operators/sequence_softmax_op.cc
+6
-4
paddle/operators/sequence_softmax_op.h
paddle/operators/sequence_softmax_op.h
+5
-1
未找到文件。
paddle/operators/reshape_op.cc
浏览文件 @
12f2b8eb
...
...
@@ -42,8 +42,7 @@ class ReshapeOp : public framework::OperatorWithKernel {
int64_t
capacity
=
std
::
accumulate
(
shape
.
begin
(),
shape
.
end
(),
1
,
std
::
multiplies
<
int
>
());
auto
*
in
=
ctx
.
Input
<
framework
::
Tensor
>
(
"X"
);
int64_t
in_size
=
framework
::
product
(
in
->
dims
());
PADDLE_ENFORCE_EQ
(
capacity
,
in_size
,
PADDLE_ENFORCE_EQ
(
capacity
,
in
->
numel
(),
"The size of Input(X) mismatches with Attr(shape)."
);
// resize output
std
::
vector
<
int64_t
>
shape_int64
(
shape
.
size
(),
0
);
...
...
paddle/operators/sequence_softmax_op.cc
浏览文件 @
12f2b8eb
...
...
@@ -30,18 +30,20 @@ class SequenceSoftmaxOp : public framework::OperatorWithKernel {
"Output(Out) of SequenceSoftmaxOp should not be null."
);
auto
*
x
=
ctx
.
Input
<
framework
::
LoDTensor
>
(
"X"
);
auto
dims
=
x
->
dims
();
auto
lod
=
x
->
lod
();
PADDLE_ENFORCE_EQ
(
lod
.
size
(),
1UL
,
"Only support one level sequence now."
);
auto
dims
=
x
->
dims
(
);
PADDLE_ENFORCE_GE
(
dims
[
0
],
/* batch_size */
static_cast
<
int64_t
>
(
lod
[
0
].
size
()
-
1
),
"The first dimension of Input(X) should be larger than batch size."
);
PADDLE_ENFORCE_EQ
(
x
->
numel
(),
static_cast
<
int64_t
>
(
lod
[
0
].
size
()
-
1
),
const
size_t
level
=
lod
.
size
()
-
1
;
PADDLE_ENFORCE_EQ
(
x
->
numel
(),
static_cast
<
int64_t
>
(
lod
[
level
].
back
()),
"The width of each timestep in Input(X) of "
"SequenceSoftmaxOp should be 1."
);
dims
[
0
]
=
lod
[
0
].
size
()
-
1
;
std
::
cout
<<
DebugString
()
<<
std
::
endl
;
ctx
.
Output
<
framework
::
LoDTensor
>
(
"Out"
)
->
Resize
({
dims
});
}
};
...
...
paddle/operators/sequence_softmax_op.h
浏览文件 @
12f2b8eb
...
...
@@ -38,7 +38,7 @@ class SequenceSoftmaxKernel : public framework::OpKernel {
auto
*
out
=
ctx
.
Output
<
LoDTensor
>
(
"Out"
);
auto
lod
=
x
->
lod
();
const
size_t
level
=
lod
.
size
();
const
size_t
level
=
lod
.
size
()
-
1
;
out
->
mutable_data
<
T
>
(
ctx
.
GetPlace
());
for
(
int
i
=
0
;
i
<
static_cast
<
int
>
(
lod
[
level
].
size
())
-
1
;
++
i
)
{
...
...
@@ -47,6 +47,10 @@ class SequenceSoftmaxKernel : public framework::OpKernel {
Tensor
x_i
=
x
->
Slice
<
T
>
(
start_pos
,
end_pos
);
Tensor
out_i
=
out
->
Slice
<
T
>
(
start_pos
,
end_pos
);
// Reshape from (end_pos - start_pos) x 1UL to 1UL x (end_pos - start_pos)
framework
::
DDim
dims
=
framework
::
make_ddim
({
1UL
,
end_pos
-
start_pos
});
x_i
.
Resize
(
dims
);
out_i
.
Resize
(
dims
);
math
::
SoftmaxFunctor
<
Place
,
T
>
()(
&
x_i
,
&
out_i
,
ctx
);
}
}
...
...
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