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
10779460
P
Paddle
项目概览
PaddlePaddle
/
Paddle
大约 1 年 前同步成功
通知
2299
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
提交
10779460
编写于
1月 10, 2018
作者:
Y
Yibing Liu
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Simplify calc in test_sequence_erase_op
上级
7b9d5b32
变更
4
隐藏空白更改
内联
并排
Showing
4 changed file
with
30 addition
and
45 deletion
+30
-45
paddle/operators/sequence_erase_op.cc
paddle/operators/sequence_erase_op.cc
+1
-1
paddle/operators/sequence_erase_op.cu
paddle/operators/sequence_erase_op.cu
+2
-0
paddle/operators/sequence_erase_op.h
paddle/operators/sequence_erase_op.h
+15
-10
python/paddle/v2/fluid/tests/test_sequence_erase_op.py
python/paddle/v2/fluid/tests/test_sequence_erase_op.py
+12
-34
未找到文件。
paddle/operators/sequence_erase_op.cc
浏览文件 @
10779460
...
@@ -50,7 +50,7 @@ class SequenceEraseOpMaker : public framework::OpProtoAndCheckerMaker {
...
@@ -50,7 +50,7 @@ class SequenceEraseOpMaker : public framework::OpProtoAndCheckerMaker {
AddComment
(
R"DOC(
AddComment
(
R"DOC(
Sequence Erase Operator.
Sequence Erase Operator.
Sequence erase operator erases tokens specified by Attr(tokens)
in
the input
Sequence erase operator erases tokens specified by Attr(tokens)
from
the input
sequences Input(X), and outputs the remaining data and modifies the LoD
sequences Input(X), and outputs the remaining data and modifies the LoD
information at the same time. For example, given a 2-D LoDTensor
information at the same time. For example, given a 2-D LoDTensor
...
...
paddle/operators/sequence_erase_op.cu
浏览文件 @
10779460
...
@@ -70,6 +70,8 @@ class SequenceEraseOpCUDAKernel : public framework::OpKernel<T> {
...
@@ -70,6 +70,8 @@ class SequenceEraseOpCUDAKernel : public framework::OpKernel<T> {
auto
lod
=
in
->
lod
();
auto
lod
=
in
->
lod
();
PADDLE_ENFORCE_EQ
(
lod
.
size
(),
1UL
,
"Only support one level sequence now."
);
PADDLE_ENFORCE_EQ
(
lod
.
size
(),
1UL
,
"Only support one level sequence now."
);
PADDLE_ENFORCE_EQ
(
lod
[
0
].
back
(),
(
size_t
)
in
->
numel
(),
"The actual size mismatches with the LoD information."
);
auto
tokens
=
ctx
.
Attr
<
std
::
vector
<
T
>>
(
"tokens"
);
auto
tokens
=
ctx
.
Attr
<
std
::
vector
<
T
>>
(
"tokens"
);
auto
tokens_len
=
tokens
.
size
();
auto
tokens_len
=
tokens
.
size
();
auto
in_len
=
in
->
numel
();
auto
in_len
=
in
->
numel
();
...
...
paddle/operators/sequence_erase_op.h
浏览文件 @
10779460
...
@@ -28,22 +28,27 @@ class SequenceEraseKernel : public framework::OpKernel<T> {
...
@@ -28,22 +28,27 @@ class SequenceEraseKernel : public framework::OpKernel<T> {
auto
lod
=
in
->
lod
();
auto
lod
=
in
->
lod
();
PADDLE_ENFORCE_EQ
(
lod
.
size
(),
1UL
,
"Only support one level sequence now."
);
PADDLE_ENFORCE_EQ
(
lod
.
size
(),
1UL
,
"Only support one level sequence now."
);
PADDLE_ENFORCE_EQ
(
lod
[
0
].
back
(),
(
size_t
)
in
->
numel
(),
"The actual size mismatches with the LoD information."
);
auto
tokens
=
ctx
.
Attr
<
std
::
vector
<
int
>>
(
"tokens"
);
auto
tokens
=
ctx
.
Attr
<
std
::
vector
<
int
>>
(
"tokens"
);
auto
in_len
=
in
->
numel
();
auto
in_len
=
in
->
numel
();
auto
in_dat
=
in
->
data
<
T
>
();
auto
in_dat
=
in
->
data
<
T
>
();
auto
lod0
=
lod
[
0
];
auto
lod0
=
lod
[
0
];
std
::
vector
<
size_t
>
num_erased
(
in_len
+
1
,
0
);
std
::
vector
<
size_t
>
num_erased
(
in_len
+
1
,
0
);
for
(
int64_t
i
=
1
;
i
<
in_len
+
1
;
++
i
)
{
std
::
vector
<
size_t
>
out_lod0
(
1
,
0
);
num_erased
[
i
]
=
num_erased
[
i
-
1
];
for
(
size_t
i
=
0
;
i
<
lod0
.
size
()
-
1
;
++
i
)
{
if
(
std
::
find
(
tokens
.
begin
(),
tokens
.
end
(),
in_dat
[
i
-
1
])
!=
size_t
num_out
=
0
;
tokens
.
end
())
{
for
(
auto
j
=
lod0
[
i
]
+
1
;
j
<=
lod0
[
i
+
1
];
++
j
)
{
num_erased
[
i
]
+=
1
;
num_erased
[
j
]
=
num_erased
[
j
-
1
];
if
(
std
::
find
(
tokens
.
begin
(),
tokens
.
end
(),
in_dat
[
j
-
1
])
!=
tokens
.
end
())
{
num_erased
[
j
]
+=
1
;
}
else
{
num_out
+=
1
;
}
}
}
}
out_lod0
.
push_back
(
out_lod0
.
back
()
+
num_out
);
std
::
vector
<
size_t
>
out_lod0
(
lod0
.
size
(),
0
);
for
(
size_t
i
=
1
;
i
<
lod0
.
size
();
++
i
)
{
out_lod0
[
i
]
=
lod0
[
i
]
-
num_erased
[
lod0
[
i
]];
}
}
auto
out_len
=
in_len
-
num_erased
[
in_len
];
auto
out_len
=
in_len
-
num_erased
[
in_len
];
...
...
python/paddle/v2/fluid/tests/test_sequence_erase_op.py
浏览文件 @
10779460
...
@@ -4,32 +4,23 @@ from op_test import OpTest
...
@@ -4,32 +4,23 @@ from op_test import OpTest
def
sequence_erase
(
in_seq
,
lod0
,
tokens
):
def
sequence_erase
(
in_seq
,
lod0
,
tokens
):
# num_erased[i]: the number of elments to be removed before #i elements
new_lod0
=
[
0
]
num_erased
=
[
0
]
*
(
len
(
in_seq
)
+
1
)
out_seq
=
[]
for
i
in
range
(
1
,
len
(
in_seq
)
+
1
):
for
i
in
range
(
0
,
len
(
lod0
)
-
1
):
num_erased
[
i
]
=
num_erased
[
i
-
1
]
num_out
=
0
if
in_seq
[
i
-
1
]
in
tokens
:
for
dat
in
in_seq
[
lod0
[
i
]:
lod0
[
i
+
1
]]:
num_erased
[
i
]
+=
1
if
dat
not
in
tokens
:
out_seq
.
append
(
dat
)
# recalculate lod information
num_out
+=
1
new_lod0
=
[
0
]
*
len
(
lod0
)
new_lod0
.
append
(
new_lod0
[
-
1
]
+
num_out
)
for
i
in
range
(
1
,
len
(
lod0
)):
return
np
.
array
(
out_seq
).
astype
(
"int32"
),
new_lod0
new_lod0
[
i
]
=
lod0
[
i
]
-
num_erased
[
lod0
[
i
]]
out_seq
=
np
.
zeros
(
(
len
(
in_seq
)
-
num_erased
[
len
(
in_seq
)],
1
)).
astype
(
"int32"
)
for
i
in
range
(
0
,
len
(
in_seq
)):
if
num_erased
[
i
]
==
num_erased
[
i
+
1
]:
out_seq
[
i
-
num_erased
[
i
]]
=
in_seq
[
i
]
# else in_seq[i] needs to be removed
return
out_seq
,
new_lod0
class
TestSequenceEraseOp
(
OpTest
):
class
TestSequenceEraseOp
(
OpTest
):
def
setUp
(
self
):
def
setUp
(
self
):
self
.
op_type
=
"sequence_erase"
self
.
op_type
=
"sequence_erase"
in_seq
=
np
.
random
.
randint
(
0
,
10
,
(
1
0
,
1
)).
astype
(
"int32"
)
in_seq
=
np
.
random
.
randint
(
0
,
10
,
(
3
0
,
1
)).
astype
(
"int32"
)
lod
=
[[
0
,
3
,
6
,
1
0
]]
lod
=
[[
0
,
9
,
13
,
24
,
3
0
]]
tokens
=
[
2
,
3
,
5
]
tokens
=
[
2
,
3
,
5
]
out_seq
,
new_lod0
=
sequence_erase
(
in_seq
,
lod
[
0
],
tokens
)
out_seq
,
new_lod0
=
sequence_erase
(
in_seq
,
lod
[
0
],
tokens
)
self
.
attrs
=
{
'tokens'
:
tokens
}
self
.
attrs
=
{
'tokens'
:
tokens
}
...
@@ -41,17 +32,4 @@ class TestSequenceEraseOp(OpTest):
...
@@ -41,17 +32,4 @@ class TestSequenceEraseOp(OpTest):
if
__name__
==
'__main__'
:
if
__name__
==
'__main__'
:
"""
in_seq = np.random.randint(0, 10, (30, 1)).astype("int32")
lod0 = [0, 5, 15, 30]
tokens = [2, 5]
out_seq, new_lod = sequence_erase(in_seq, lod0, tokens)
print lod0, new_lod
print("compare")
for i in range(0, len(lod0)-1):
print(np.transpose(in_seq[lod0[i] : lod0[i+1]]))
print(np.transpose(out_seq[new_lod[i] : new_lod[i+1]]))
print("
\n
")
"""
unittest
.
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