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
csdn_franckjun
Paddle
提交
5a159f34
P
Paddle
项目概览
csdn_franckjun
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
1
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1
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
1
Issue
1
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 >>
未验证
提交
5a159f34
编写于
3月 14, 2018
作者:
C
chengduo
提交者:
GitHub
3月 14, 2018
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #8934 from chengduoZH/feature/Enhance_regularizer_py
Enhance regularizer.py
上级
ba65d54d
93107ce1
变更
2
显示空白变更内容
内联
并排
Showing
2 changed file
with
38 addition
and
6 deletion
+38
-6
python/paddle/fluid/regularizer.py
python/paddle/fluid/regularizer.py
+34
-5
python/paddle/fluid/tests/book/test_machine_translation.py
python/paddle/fluid/tests/book/test_machine_translation.py
+4
-1
未找到文件。
python/paddle/fluid/regularizer.py
浏览文件 @
5a159f34
...
...
@@ -13,6 +13,7 @@
# limitations under the License.
import
framework
from
.
import
core
__all__
=
[
'append_regularization_ops'
,
...
...
@@ -46,9 +47,9 @@ def append_regularization_ops(parameters_and_grads, regularization=None):
regularization_term
=
None
if
param
.
regularizer
is
not
None
:
# Add variable for regularization term in grad block
regularization_term
=
param
.
regularizer
(
param
,
grad
.
block
)
regularization_term
=
param
.
regularizer
(
param
,
grad
,
grad
.
block
)
elif
regularization
is
not
None
:
regularization_term
=
regularization
(
param
,
grad
.
block
)
regularization_term
=
regularization
(
param
,
grad
,
grad
.
block
)
# If no gradient or no regularization specified,
# then we don't need to do anything
...
...
@@ -82,7 +83,7 @@ class WeightDecayRegularizer(object):
def
__init__
(
self
):
pass
def
__call__
(
self
,
param
,
block
):
def
__call__
(
self
,
param
,
grad
,
block
):
"""Add corresponding weight decay operations to the network
"""
raise
NotImplementedError
()
...
...
@@ -102,7 +103,7 @@ class L2DecayRegularizer(WeightDecayRegularizer):
super
(
L2DecayRegularizer
,
self
).
__init__
()
self
.
_regularization_coeff
=
regularization_coeff
def
__call__
(
self
,
param
,
block
):
def
__call__
(
self
,
param
,
grad
,
block
):
"""Add L2 weight decay ops to network
Adds L2 weight decay ops.
...
...
@@ -117,8 +118,23 @@ class L2DecayRegularizer(WeightDecayRegularizer):
"""
assert
isinstance
(
param
,
framework
.
Parameter
)
assert
isinstance
(
block
,
framework
.
Block
)
decay
=
block
.
create_var
(
dtype
=
"float32"
,
shape
=
param
.
shape
,
lod_level
=
param
.
lod_level
)
if
grad
.
type
==
core
.
VarDesc
.
VarType
.
SELECTED_ROWS
:
decay
=
block
.
create_var
(
dtype
=
"float32"
,
shape
=
param
.
shape
,
type
=
core
.
VarDesc
.
VarType
.
SELECTED_ROWS
)
block
.
append_op
(
type
=
'lookup_table'
,
inputs
=
{
'W'
:
param
,
'Ids'
:
grad
},
outputs
=
{
'Out'
:
decay
},
attrs
=
{
'is_sparse'
:
True
})
param
=
decay
# Append Op to calculate decay
block
.
append_op
(
type
=
'scale'
,
...
...
@@ -141,7 +157,7 @@ class L1DecayRegularizer(WeightDecayRegularizer):
super
(
L1DecayRegularizer
,
self
).
__init__
()
self
.
_regularization_coeff
=
regularization_coeff
def
__call__
(
self
,
param
,
block
):
def
__call__
(
self
,
param
,
grad
,
block
):
"""Add L1 weight decay ops to network
Adds L1 weight decay ops.
...
...
@@ -158,6 +174,19 @@ class L1DecayRegularizer(WeightDecayRegularizer):
assert
isinstance
(
block
,
framework
.
Block
)
decay
=
block
.
create_var
(
dtype
=
"float32"
,
shape
=
param
.
shape
,
lod_level
=
param
.
lod_level
)
if
grad
.
type
==
core
.
VarDesc
.
VarType
.
SELECTED_ROWS
:
decay
=
block
.
create_var
(
dtype
=
"float32"
,
shape
=
param
.
shape
,
type
=
core
.
VarDesc
.
VarType
.
SELECTED_ROWS
)
block
.
append_op
(
type
=
'lookup_table'
,
inputs
=
{
'W'
:
param
,
'Ids'
:
grad
},
outputs
=
{
'Out'
:
decay
},
attrs
=
{
'is_sparse'
:
True
})
# Append sign op
block
.
append_op
(
type
=
'sign'
,
inputs
=
{
"X"
:
param
},
outputs
=
{
"Out"
:
decay
})
...
...
python/paddle/fluid/tests/book/test_machine_translation.py
浏览文件 @
5a159f34
...
...
@@ -181,7 +181,10 @@ def train_main(use_cuda, is_sparse, is_local=True):
cost
=
pd
.
cross_entropy
(
input
=
rnn_out
,
label
=
label
)
avg_cost
=
pd
.
mean
(
cost
)
optimizer
=
fluid
.
optimizer
.
Adagrad
(
learning_rate
=
1e-4
)
optimizer
=
fluid
.
optimizer
.
Adagrad
(
learning_rate
=
1e-4
,
regularization
=
fluid
.
regularizer
.
L2DecayRegularizer
(
regularization_coeff
=
0.1
))
optimize_ops
,
params_grads
=
optimizer
.
minimize
(
avg_cost
)
train_data
=
paddle
.
batch
(
...
...
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