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
afeixing77
Paddle
提交
d1de7ec6
P
Paddle
项目概览
afeixing77
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
0
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
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
0
Issue
0
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
d1de7ec6
编写于
10月 02, 2017
作者:
K
Kexin Zhao
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Change learning rate from attribute to input tensor
上级
05cbd4da
变更
3
隐藏空白更改
内联
并排
Showing
3 changed file
with
16 addition
and
5 deletion
+16
-5
paddle/operators/adagrad_op.cc
paddle/operators/adagrad_op.cc
+6
-1
paddle/operators/adagrad_op.h
paddle/operators/adagrad_op.h
+8
-1
python/paddle/v2/framework/tests/test_adagrad_op.py
python/paddle/v2/framework/tests/test_adagrad_op.py
+2
-3
未找到文件。
paddle/operators/adagrad_op.cc
浏览文件 @
d1de7ec6
...
@@ -29,12 +29,17 @@ class AdagradOp : public framework::OperatorWithKernel {
...
@@ -29,12 +29,17 @@ class AdagradOp : public framework::OperatorWithKernel {
"Input(grad) of AdagradOp should not be null."
);
"Input(grad) of AdagradOp should not be null."
);
PADDLE_ENFORCE
(
ctx
->
HasInput
(
"moment"
),
PADDLE_ENFORCE
(
ctx
->
HasInput
(
"moment"
),
"Input(moment) of AdagradOp should not be null."
);
"Input(moment) of AdagradOp should not be null."
);
PADDLE_ENFORCE
(
ctx
->
HasInput
(
"learning_rate"
),
"Input(learning_rate) of AdagradOp should not be null."
);
PADDLE_ENFORCE
(
ctx
->
HasOutput
(
"param_out"
),
PADDLE_ENFORCE
(
ctx
->
HasOutput
(
"param_out"
),
"Output(param_out) of AdagradOp should not be null."
);
"Output(param_out) of AdagradOp should not be null."
);
PADDLE_ENFORCE
(
ctx
->
HasOutput
(
"moment_out"
),
PADDLE_ENFORCE
(
ctx
->
HasOutput
(
"moment_out"
),
"Output(moment_out) of AdagradOp should not be null."
);
"Output(moment_out) of AdagradOp should not be null."
);
auto
lr_dims
=
ctx
->
GetInputDim
(
"learning_rate"
);
PADDLE_ENFORCE_EQ
(
framework
::
product
(
lr_dims
),
1
,
"learning_rate should have one element"
);
auto
param_dim
=
ctx
->
GetInputDim
(
"param"
);
auto
param_dim
=
ctx
->
GetInputDim
(
"param"
);
PADDLE_ENFORCE_EQ
(
PADDLE_ENFORCE_EQ
(
param_dim
,
ctx
->
GetInputDim
(
"grad"
),
param_dim
,
ctx
->
GetInputDim
(
"grad"
),
...
@@ -56,11 +61,11 @@ class AdagradOpMaker : public framework::OpProtoAndCheckerMaker {
...
@@ -56,11 +61,11 @@ class AdagradOpMaker : public framework::OpProtoAndCheckerMaker {
AddInput
(
"param"
,
"Input parameter"
);
AddInput
(
"param"
,
"Input parameter"
);
AddInput
(
"grad"
,
"Input gradient"
);
AddInput
(
"grad"
,
"Input gradient"
);
AddInput
(
"moment"
,
"Second moment"
);
AddInput
(
"moment"
,
"Second moment"
);
AddInput
(
"learning_rate"
,
"learning rate of adagrad"
);
AddOutput
(
"param_out"
,
"Output parameter"
);
AddOutput
(
"param_out"
,
"Output parameter"
);
AddOutput
(
"moment_out"
,
"Output second moment"
);
AddOutput
(
"moment_out"
,
"Output second moment"
);
AddAttr
<
float
>
(
"learning_rate"
,
"Learning rate"
);
AddAttr
<
float
>
(
"epsilon"
,
"Constant for numerical stability"
);
AddAttr
<
float
>
(
"epsilon"
,
"Constant for numerical stability"
);
AddComment
(
R"DOC(
AddComment
(
R"DOC(
...
...
paddle/operators/adagrad_op.h
浏览文件 @
d1de7ec6
...
@@ -20,6 +20,11 @@ namespace paddle {
...
@@ -20,6 +20,11 @@ namespace paddle {
namespace
operators
{
namespace
operators
{
using
Tensor
=
framework
::
Tensor
;
using
Tensor
=
framework
::
Tensor
;
template
<
typename
T
,
int
MajorType
=
Eigen
::
RowMajor
,
typename
IndexType
=
Eigen
::
DenseIndex
>
using
EigenScalar
=
framework
::
EigenScalar
<
T
,
MajorType
,
IndexType
>
;
template
<
typename
T
,
int
MajorType
=
Eigen
::
RowMajor
,
template
<
typename
T
,
int
MajorType
=
Eigen
::
RowMajor
,
typename
IndexType
=
Eigen
::
DenseIndex
>
typename
IndexType
=
Eigen
::
DenseIndex
>
using
EigenVector
=
framework
::
EigenVector
<
T
,
MajorType
,
IndexType
>
;
using
EigenVector
=
framework
::
EigenVector
<
T
,
MajorType
,
IndexType
>
;
...
@@ -34,12 +39,14 @@ class AdagradOpKernel : public framework::OpKernel<T> {
...
@@ -34,12 +39,14 @@ class AdagradOpKernel : public framework::OpKernel<T> {
param_out
->
mutable_data
<
T
>
(
ctx
.
GetPlace
());
param_out
->
mutable_data
<
T
>
(
ctx
.
GetPlace
());
moment_out
->
mutable_data
<
T
>
(
ctx
.
GetPlace
());
moment_out
->
mutable_data
<
T
>
(
ctx
.
GetPlace
());
float
lr
=
ctx
.
Attr
<
float
>
(
"learning_rate"
)
;
float
lr
=
ctx
.
Input
<
Tensor
>
(
"learning_rate"
)
->
data
<
float
>
()[
0
]
;
float
epsilon
=
ctx
.
Attr
<
float
>
(
"epsilon"
);
float
epsilon
=
ctx
.
Attr
<
float
>
(
"epsilon"
);
auto
p
=
EigenVector
<
T
>::
Flatten
(
*
ctx
.
Input
<
Tensor
>
(
"param"
));
auto
p
=
EigenVector
<
T
>::
Flatten
(
*
ctx
.
Input
<
Tensor
>
(
"param"
));
auto
g
=
EigenVector
<
T
>::
Flatten
(
*
ctx
.
Input
<
Tensor
>
(
"grad"
));
auto
g
=
EigenVector
<
T
>::
Flatten
(
*
ctx
.
Input
<
Tensor
>
(
"grad"
));
auto
m
=
EigenVector
<
T
>::
Flatten
(
*
ctx
.
Input
<
Tensor
>
(
"moment"
));
auto
m
=
EigenVector
<
T
>::
Flatten
(
*
ctx
.
Input
<
Tensor
>
(
"moment"
));
auto
lr
=
EigenScalar
<
T
>::
From
(
*
ctx
.
Input
<
Tensor
>
(
"learning_rate"
));
auto
p_out
=
EigenVector
<
T
>::
Flatten
(
*
param_out
);
auto
p_out
=
EigenVector
<
T
>::
Flatten
(
*
param_out
);
auto
m_out
=
EigenVector
<
T
>::
Flatten
(
*
moment_out
);
auto
m_out
=
EigenVector
<
T
>::
Flatten
(
*
moment_out
);
auto
place
=
ctx
.
GetEigenDevice
<
Place
>
();
auto
place
=
ctx
.
GetEigenDevice
<
Place
>
();
...
...
python/paddle/v2/framework/tests/test_adagrad_op.py
浏览文件 @
d1de7ec6
...
@@ -11,7 +11,7 @@ class TestAdagradOp(OpTest):
...
@@ -11,7 +11,7 @@ class TestAdagradOp(OpTest):
grad
=
np
.
random
.
random
((
123
,
321
)).
astype
(
"float32"
)
grad
=
np
.
random
.
random
((
123
,
321
)).
astype
(
"float32"
)
moment
=
np
.
zeros
((
123
,
321
)).
astype
(
"float32"
)
moment
=
np
.
zeros
((
123
,
321
)).
astype
(
"float32"
)
l
earning_rate
=
0.01
l
r
=
np
.
array
([
0.01
]).
astype
(
"float32"
)
epsilon
=
1e-6
epsilon
=
1e-6
self
.
inputs
=
{
'param'
:
param
,
'grad'
:
grad
,
'moment'
:
moment
}
self
.
inputs
=
{
'param'
:
param
,
'grad'
:
grad
,
'moment'
:
moment
}
...
@@ -19,8 +19,7 @@ class TestAdagradOp(OpTest):
...
@@ -19,8 +19,7 @@ class TestAdagradOp(OpTest):
self
.
attrs
=
{
'learning_rate'
:
learning_rate
,
'epsilon'
:
epsilon
}
self
.
attrs
=
{
'learning_rate'
:
learning_rate
,
'epsilon'
:
epsilon
}
moment_out
=
moment
+
grad
*
grad
moment_out
=
moment
+
grad
*
grad
param_out
=
param
-
learning_rate
*
grad
/
(
np
.
sqrt
(
moment_out
)
+
param_out
=
param
-
lr
*
grad
/
(
np
.
sqrt
(
moment_out
)
+
epsilon
)
epsilon
)
self
.
outputs
=
{
'param_out'
:
param_out
,
'moment_out'
:
moment_out
}
self
.
outputs
=
{
'param_out'
:
param_out
,
'moment_out'
:
moment_out
}
...
...
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