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
wzpei9
Paddle
提交
680aec21
P
Paddle
项目概览
wzpei9
/
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 >>
提交
680aec21
编写于
1月 20, 2018
作者:
W
wanghaoshuang
浏览文件
操作
浏览文件
下载
差异文件
Merge branch 'develop' of
https://github.com/PaddlePaddle/Paddle
into ctc_evaluator_py
上级
25dec82f
8e08b0a2
变更
14
隐藏空白更改
内联
并排
Showing
14 changed file
with
626 addition
and
57 deletion
+626
-57
doc/api/v2/fluid/layers.rst
doc/api/v2/fluid/layers.rst
+5
-0
paddle/framework/operator.cc
paddle/framework/operator.cc
+7
-3
paddle/operators/recv_op.cc
paddle/operators/recv_op.cc
+7
-10
paddle/operators/scale_op.cc
paddle/operators/scale_op.cc
+1
-1
paddle/operators/send_recv_op_test.cc
paddle/operators/send_recv_op_test.cc
+1
-4
paddle/operators/sequence_reshape_op.cc
paddle/operators/sequence_reshape_op.cc
+130
-0
paddle/operators/sequence_reshape_op.cu
paddle/operators/sequence_reshape_op.cu
+30
-0
paddle/operators/sequence_reshape_op.h
paddle/operators/sequence_reshape_op.h
+86
-0
python/paddle/v2/fluid/distribute_transpiler.py
python/paddle/v2/fluid/distribute_transpiler.py
+1
-1
python/paddle/v2/fluid/distribute_transpiler_simple.py
python/paddle/v2/fluid/distribute_transpiler_simple.py
+1
-1
python/paddle/v2/fluid/layers/nn.py
python/paddle/v2/fluid/layers/nn.py
+92
-37
python/paddle/v2/fluid/tests/book_distribute/notest_dist_image_classification.py
...tests/book_distribute/notest_dist_image_classification.py
+173
-0
python/paddle/v2/fluid/tests/test_layers.py
python/paddle/v2/fluid/tests/test_layers.py
+8
-0
python/paddle/v2/fluid/tests/test_sequence_reshape.py
python/paddle/v2/fluid/tests/test_sequence_reshape.py
+84
-0
未找到文件。
doc/api/v2/fluid/layers.rst
浏览文件 @
680aec21
...
...
@@ -514,3 +514,8 @@ l2_normalize
------------
.. autofunction:: paddle.v2.fluid.layers.l2_normalize
:noindex:
sequence_reshape
----------------
.. autofunction:: paddle.v2.fluid.layers.sequence_reshape
:noindex:
paddle/framework/operator.cc
浏览文件 @
680aec21
...
...
@@ -485,9 +485,15 @@ void OperatorWithKernel::Run(const Scope& scope,
// }
auto
expected_kernel_key
=
this
->
GetExpectedKernelType
(
ctx
);
VLOG
(
3
)
<<
"expected_kernel_key:"
<<
expected_kernel_key
;
auto
kernel_iter
=
kernels
.
find
(
expected_kernel_key
);
if
(
kernel_iter
==
kernels
.
end
())
{
PADDLE_THROW
(
"op %s does not have kernel for %s"
,
type_
,
KernelTypeToString
(
expected_kernel_key
));
}
// do data transform
Scope
&
new_scope
=
scope
.
NewScope
();
for
(
auto
&
var_name_item
:
this
->
Inputs
())
{
...
...
@@ -520,8 +526,6 @@ void OperatorWithKernel::Run(const Scope& scope,
}
}
auto
kernel_iter
=
kernels
.
find
(
expected_kernel_key
);
auto
*
new_dev_ctx
=
pool
.
Get
(
expected_kernel_key
.
place_
);
kernel_iter
->
second
->
Compute
(
ExecutionContext
(
*
this
,
new_scope
,
*
new_dev_ctx
));
...
...
paddle/operators/recv_op.cc
浏览文件 @
680aec21
...
...
@@ -34,9 +34,7 @@ limitations under the License. */
namespace
paddle
{
namespace
operators
{
constexpr
int
kCondStart
=
0
;
constexpr
int
kCondRunning
=
1
;
constexpr
int
kCondDone
=
2
;
constexpr
char
kOptimizeBlock
[]
=
"OptimizeBlock"
;
void
RunServer
(
std
::
shared_ptr
<
detail
::
AsyncGRPCServer
>
service
)
{
service
->
RunSyncUpdate
();
...
...
@@ -99,10 +97,8 @@ class RecvOp : public framework::OperatorBase {
auto
fan_in
=
Attr
<
int
>
(
"Fanin"
);
size_t
param_count
=
param_list
.
size
();
std
::
string
program_str
=
Attr
<
std
::
string
>
(
"OptimizeProgram"
);
framework
::
proto
::
ProgramDesc
program_desc
;
program_desc
.
ParseFromString
(
program_str
);
framework
::
ProgramDesc
program
(
program_desc
);
auto
*
block
=
Attr
<
framework
::
BlockDesc
*>
(
kOptimizeBlock
);
auto
*
program
=
block
->
Program
();
framework
::
Executor
executor
(
dev_place
);
// TODO(typhoonzero): change this to a while_op for every cluster-batch.
...
...
@@ -142,8 +138,9 @@ class RecvOp : public framework::OperatorBase {
if
(
exit_flag
)
{
break
;
}
try
{
executor
.
Run
(
program
,
&
recv_scope
,
0
,
/*global_block*/
executor
.
Run
(
*
program
,
&
recv_scope
,
block
->
ID
()
,
/*global_block*/
false
/*create_local_scope*/
,
false
/*create_vars*/
);
}
catch
(
std
::
exception
&
e
)
{
LOG
(
ERROR
)
<<
"run sub program error "
<<
e
.
what
();
...
...
@@ -175,8 +172,8 @@ This operator will recv tensor from send_op
"IP address to listen on."
)
.
SetDefault
(
"127.0.0.1:6164"
)
.
AddCustomChecker
([](
const
std
::
string
&
ip
)
{
return
!
ip
.
empty
();
});
AddAttr
<
std
::
string
>
(
"OptimizeProgram"
,
"type string"
,
"Serialized ProgramDesc string for recv to run."
);
AddAttr
<
framework
::
BlockDesc
*>
(
kOptimizeBlock
,
"Serialized ProgramDesc string for recv to run."
);
AddAttr
<
std
::
vector
<
std
::
string
>>
(
"ParamList"
,
"type list of string"
,
"grad->param name mapping to find which param to optimize."
)
...
...
paddle/operators/scale_op.cc
浏览文件 @
680aec21
...
...
@@ -48,7 +48,7 @@ Scale operator
$$Out = scale*X$$
)DOC"
);
AddAttr
<
AttrType
>
(
"scale"
,
"(float, default 0)"
"(float, default
1.
0)"
"The scaling factor of the scale operator."
)
.
SetDefault
(
1.0
);
}
...
...
paddle/operators/send_recv_op_test.cc
浏览文件 @
680aec21
...
...
@@ -130,10 +130,7 @@ void StartServerNet(bool is_sparse) {
attrs
.
insert
({
"endpoint"
,
std
::
string
(
"127.0.0.1:6174"
)});
attrs
.
insert
({
"ParamList"
,
std
::
vector
<
std
::
string
>
({
"Out"
})});
attrs
.
insert
({
"GradList"
,
std
::
vector
<
std
::
string
>
({
"x1"
})});
std
::
string
program_proto
;
PADDLE_ENFORCE
(
program
.
Proto
()
->
SerializeToString
(
&
program_proto
));
attrs
.
insert
({
"OptimizeProgram"
,
program_proto
});
attrs
.
insert
({
"OptimizeBlock"
,
block
});
recv_op
=
f
::
OpRegistry
::
CreateOp
(
"recv"
,
{{
"RX"
,
{
"x1"
}}},
{},
attrs
);
recv_op
->
Run
(
scope
,
place
);
}
...
...
paddle/operators/sequence_reshape_op.cc
0 → 100644
浏览文件 @
680aec21
/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/operators/sequence_reshape_op.h"
#include "paddle/framework/ddim.h"
namespace
paddle
{
namespace
operators
{
class
SequenceReshapeOp
:
public
framework
::
OperatorWithKernel
{
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
void
InferShape
(
framework
::
InferShapeContext
*
ctx
)
const
override
{
PADDLE_ENFORCE
(
ctx
->
HasInput
(
"X"
),
"Input(X) of SequenceReshapeOp should not be null."
);
PADDLE_ENFORCE
(
ctx
->
HasOutput
(
"Out"
),
"Output(Out) of SequenceReshapeOp should not be null."
);
auto
x_dims
=
ctx
->
GetInputDim
(
"X"
);
auto
x_numel
=
product
(
x_dims
);
PADDLE_ENFORCE_EQ
(
x_dims
.
size
(),
2U
,
"Rank of Input(X) should be 2."
);
int
new_dim
=
ctx
->
Attrs
().
Get
<
int
>
(
"new_dim"
);
ctx
->
SetOutputDim
(
"Out"
,
{
x_numel
/
new_dim
,
static_cast
<
int64_t
>
(
new_dim
)});
}
};
class
SequenceReshapeOpMaker
:
public
framework
::
OpProtoAndCheckerMaker
{
public:
SequenceReshapeOpMaker
(
OpProto
*
proto
,
OpAttrChecker
*
op_checker
)
:
OpProtoAndCheckerMaker
(
proto
,
op_checker
)
{
AddInput
(
"X"
,
"(LoDTensor, default LoDTensor<float>) A 2-D LoDTensor with shape "
"being [N, M]."
);
AddOutput
(
"Out"
,
"(LoDTensor, default LoDTensor<float>) A 2-D LoDTensor with "
"shape [T, new_dim] where T is calculated based on X.lod, M and "
"new_dim."
);
AddAttr
<
int
>
(
"new_dim"
,
"Sequence dimension of the output LoDTensor."
);
AddComment
(
R"DOC(
Sequence Reshape Operator.
This operator will rearrange the input sequences. The new dimension is set by
attribute and length of each sequence may change longer or shorter which is
decided by original length, original dimension and new dimension. The following
example will help to illustrate the function of this operator:
x is a LoDTensor:
x.lod = [[0, 2, 6]]
x.data = [[1, 2], [3, 4],
[5, 6], [7, 8], [9, 10], [11, 12]]
x.dims = [6, 2]
set new_dim = 4
then out is a LoDTensor:
out.lod = [[0, 1, 3]]
out.data = [[1, 2, 3, 4],
[5, 6, 7, 8], [9, 10, 11, 12]]
out.dims = [3, 4]
Currently, only 1-level LoDTensor is supported and please make sure (original
length * original dimension) can be divided by new_dim with no remainder for
each sequence.
)DOC"
);
}
};
class
SequenceReshapeGradOp
:
public
framework
::
OperatorWithKernel
{
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
void
InferShape
(
framework
::
InferShapeContext
*
ctx
)
const
override
{
PADDLE_ENFORCE
(
ctx
->
HasInput
(
framework
::
GradVarName
(
"Out"
)),
"Input(Out@GRAD) of SequenceReshapeGradOp should not be null."
);
PADDLE_ENFORCE
(
ctx
->
HasInput
(
"X"
),
"Input(X) of SequenceReshapeGradOp should not be null."
);
ctx
->
SetOutputDim
(
framework
::
GradVarName
(
"X"
),
ctx
->
GetInputDim
(
"X"
));
ctx
->
ShareLoD
(
"X"
,
/*->*/
framework
::
GradVarName
(
"X"
));
}
};
class
SequenceReshapeGradOpMaker
:
public
framework
::
SingleGradOpDescMaker
{
public:
using
framework
::
SingleGradOpDescMaker
::
SingleGradOpDescMaker
;
protected:
std
::
unique_ptr
<
framework
::
OpDesc
>
Apply
()
const
override
{
auto
*
op_desc_ptr
=
new
framework
::
OpDesc
();
op_desc_ptr
->
SetType
(
"sequence_reshape_grad"
);
op_desc_ptr
->
SetInput
(
"X"
,
Input
(
"X"
));
op_desc_ptr
->
SetInput
(
framework
::
GradVarName
(
"Out"
),
OutputGrad
(
"Out"
));
op_desc_ptr
->
SetOutput
(
framework
::
GradVarName
(
"X"
),
InputGrad
(
"X"
));
op_desc_ptr
->
SetAttrMap
(
Attrs
());
return
std
::
unique_ptr
<
framework
::
OpDesc
>
(
op_desc_ptr
);
}
};
}
// namespace operators
}
// namespace paddle
namespace
ops
=
paddle
::
operators
;
REGISTER_OPERATOR
(
sequence_reshape
,
ops
::
SequenceReshapeOp
,
ops
::
SequenceReshapeOpMaker
,
ops
::
SequenceReshapeGradOpMaker
);
REGISTER_OPERATOR
(
sequence_reshape_grad
,
ops
::
SequenceReshapeGradOp
);
REGISTER_OP_CPU_KERNEL
(
sequence_reshape
,
ops
::
SequenceReshapeKernel
<
paddle
::
platform
::
CPUDeviceContext
,
float
>
,
ops
::
SequenceReshapeKernel
<
paddle
::
platform
::
CPUDeviceContext
,
double
>
,
ops
::
SequenceReshapeKernel
<
paddle
::
platform
::
CPUDeviceContext
,
int
>
,
ops
::
SequenceReshapeKernel
<
paddle
::
platform
::
CPUDeviceContext
,
int64_t
>
);
REGISTER_OP_CPU_KERNEL
(
sequence_reshape_grad
,
ops
::
SequenceReshapeGradKernel
<
paddle
::
platform
::
CPUDeviceContext
,
float
>
,
ops
::
SequenceReshapeGradKernel
<
paddle
::
platform
::
CPUDeviceContext
,
double
>
,
ops
::
SequenceReshapeGradKernel
<
paddle
::
platform
::
CPUDeviceContext
,
int64_t
>
,
ops
::
SequenceReshapeGradKernel
<
paddle
::
platform
::
CPUDeviceContext
,
int
>
);
paddle/operators/sequence_reshape_op.cu
0 → 100644
浏览文件 @
680aec21
/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/operators/sequence_reshape_op.h"
namespace
ops
=
paddle
::
operators
;
REGISTER_OP_CUDA_KERNEL
(
sequence_reshape
,
ops
::
SequenceReshapeKernel
<
paddle
::
platform
::
CUDADeviceContext
,
float
>
,
ops
::
SequenceReshapeKernel
<
paddle
::
platform
::
CUDADeviceContext
,
double
>
,
ops
::
SequenceReshapeKernel
<
paddle
::
platform
::
CUDADeviceContext
,
int
>
,
ops
::
SequenceReshapeKernel
<
paddle
::
platform
::
CUDADeviceContext
,
int64_t
>
);
REGISTER_OP_CUDA_KERNEL
(
sequence_reshape_grad
,
ops
::
SequenceReshapeGradKernel
<
paddle
::
platform
::
CUDADeviceContext
,
float
>
,
ops
::
SequenceReshapeGradKernel
<
paddle
::
platform
::
CUDADeviceContext
,
double
>
,
ops
::
SequenceReshapeGradKernel
<
paddle
::
platform
::
CUDADeviceContext
,
int64_t
>
,
ops
::
SequenceReshapeGradKernel
<
paddle
::
platform
::
CUDADeviceContext
,
int
>
);
paddle/operators/sequence_reshape_op.h
0 → 100644
浏览文件 @
680aec21
/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#pragma once
#include "paddle/framework/op_registry.h"
#include "paddle/operators/math/math_function.h"
namespace
paddle
{
namespace
operators
{
using
LoDTensor
=
framework
::
LoDTensor
;
template
<
typename
DeviceContext
,
typename
T
>
class
SequenceReshapeKernel
:
public
framework
::
OpKernel
<
T
>
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
context
)
const
override
{
auto
*
in
=
context
.
Input
<
LoDTensor
>
(
"X"
);
auto
*
out
=
context
.
Output
<
LoDTensor
>
(
"Out"
);
int
out_width
=
context
.
Attr
<
int
>
(
"new_dim"
);
auto
in_dims
=
in
->
dims
();
int64_t
in_width
=
in_dims
[
1
];
auto
&
in_lod
=
in
->
lod
();
PADDLE_ENFORCE_EQ
(
in_lod
.
size
(),
1UL
,
"Only support one level sequence now."
);
PADDLE_ENFORCE_EQ
(
in_dims
[
0
],
in_lod
[
0
].
back
(),
"Inconsistent size between X.shape[0] and X.lod()[0].back()."
);
auto
in_lod_l0
=
in_lod
[
0
];
int
seq_num
=
in_lod_l0
.
size
()
-
1
;
if
(
in_width
==
out_width
)
{
out
->
set_lod
(
in
->
lod
());
}
else
{
auto
&
out_lod
=
*
out
->
mutable_lod
();
out_lod
.
resize
(
1
);
out_lod
[
0
].
resize
(
seq_num
+
1
);
out_lod
[
0
][
0
]
=
0
;
for
(
int
i
=
0
;
i
<
seq_num
;
++
i
)
{
size_t
seq_len
=
in_lod_l0
[
i
+
1
]
-
in_lod_l0
[
i
];
size_t
offset
=
0
;
offset
=
(
seq_len
*
in_width
)
/
out_width
;
PADDLE_ENFORCE_EQ
(
offset
*
out_width
,
seq_len
*
in_width
,
"Please make sure (sequence_length * dimension) can "
"be divided by new_dim with no remainder for each "
"sequence. The %dth sequence is invalid."
,
i
+
1
);
out_lod
[
0
][
i
+
1
]
=
out_lod
[
0
][
i
]
+
offset
;
}
}
framework
::
Copy
(
*
in
,
context
.
GetPlace
(),
out
);
out
->
Resize
({
static_cast
<
int64_t
>
(
out
->
lod
()[
0
].
back
()),
out_width
});
}
};
template
<
typename
DeviceContext
,
typename
T
>
class
SequenceReshapeGradKernel
:
public
framework
::
OpKernel
<
T
>
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
context
)
const
override
{
auto
*
x_tensor_ptr
=
context
.
Input
<
LoDTensor
>
(
"X"
);
auto
*
outg_tensor_ptr
=
context
.
Input
<
LoDTensor
>
(
framework
::
GradVarName
(
"Out"
));
auto
*
xg_tensor_ptr
=
context
.
Output
<
LoDTensor
>
(
framework
::
GradVarName
(
"X"
));
xg_tensor_ptr
->
mutable_data
<
T
>
(
context
.
GetPlace
());
framework
::
Copy
(
*
outg_tensor_ptr
,
context
.
GetPlace
(),
xg_tensor_ptr
);
xg_tensor_ptr
->
Resize
(
x_tensor_ptr
->
dims
());
}
};
}
// namespace operators
}
// namespace paddle
python/paddle/v2/fluid/distribute_transpiler.py
浏览文件 @
680aec21
...
...
@@ -452,7 +452,7 @@ class DistributeTranspiler:
},
# grads to recv
outputs
=
{},
attrs
=
{
"Optimize
Program"
:
optimize_sub_program
.
desc
,
"Optimize
Block"
:
optimize_sub_program
.
global_block
()
,
"endpoint"
:
endpoint
,
"ParamList"
:
[
p
.
name
...
...
python/paddle/v2/fluid/distribute_transpiler_simple.py
浏览文件 @
680aec21
...
...
@@ -243,7 +243,7 @@ class SimpleDistributeTranspiler:
self
.
param_grad_map
[
endpoint
][
"grads"
]},
# grads to recv
outputs
=
{},
attrs
=
{
"Optimize
Program"
:
optimize_sub_program
.
desc
,
"Optimize
Block"
:
optimize_sub_program
.
global_block
()
,
"endpoint"
:
endpoint
,
"ParamList"
:
[
p
.
name
for
p
in
self
.
param_grad_map
[
endpoint
][
"params"
]],
...
...
python/paddle/v2/fluid/layers/nn.py
浏览文件 @
680aec21
...
...
@@ -28,7 +28,8 @@ __all__ = [
'batch_norm'
,
'beam_search_decode'
,
'conv2d_transpose'
,
'sequence_expand'
,
'lstm_unit'
,
'reduce_sum'
,
'reduce_mean'
,
'reduce_max'
,
'reduce_min'
,
'sequence_first_step'
,
'sequence_last_step'
,
'dropout'
,
'split'
,
'ctc_greedy_decoder'
,
'edit_distance'
,
'l2_normalize'
,
'matmul'
,
'warpctc'
'ctc_greedy_decoder'
,
'edit_distance'
,
'l2_normalize'
,
'matmul'
,
'warpctc'
,
'sequence_reshape'
]
...
...
@@ -213,33 +214,33 @@ def dynamic_lstm(input,
(https://arxiv.org/pdf/1402.1128.pdf), the formula is as follows:
.. math::
i_t & = \sigma(W_{ix}x_{t} + W_{ih}h_{t-1} + W_{ic}c_{t-1} + b_i)
f_t & = \sigma(W_{fx}x_{t} + W_{fh}h_{t-1} + W_{fc}c_{t-1} + b_f)
i_t & = \sigma(W_{ix}x_{t} + W_{ih}h_{t-1} + W_{ic}c_{t-1} + b_i)
\\
tilde{c_t} & = act_g(W_{cx}x_t + W_{ch}h_{t-1} + b_c)
f_t & = \sigma(W_{fx}x_{t} + W_{fh}h_{t-1} + W_{fc}c_{t-1} + b_f)
o_t & = \sigma(W_{ox}x_{t} + W_{oh}h_{t-1} + W_{oc}c_t + b_o)
\\
tilde{c_t} & = act_g(W_{cx}x_t + W_{ch}h_{t-1} + b_c)
c_t & = f_t \odot c_{t-1} + i_t \odot
\\
tilde{c_t}
o_t & = \sigma(W_{ox}x_{t} + W_{oh}h_{t-1} + W_{oc}c_t + b_o)
c_t & = f_t \odot c_{t-1} + i_t \odot
\\
tilde{c_t}
h_t & = o_t \odot act_h(c_t)
where the :math:`W` terms denote weight matrices (e.g. :math:`W_{xi}` is
where the :math:`W` terms denote weight matrices (e.g. :math:`W_{xi}` is
the matrix of weights from the input gate to the input), :math:`W_{ic},
\
W_{fc}, W_{oc}` are diagonal weight matrices for peephole connections. In
our implementation, we use vectors to reprenset these diagonal weight
matrices. The :math:`b` terms denote bias vectors (:math:`b_i` is the input
gate bias vector), :math:`\sigma` is the non-line activations, such as
logistic sigmoid function, and :math:`i, f, o` and :math:`c` are the input
gate, forget gate, output gate, and cell activation vectors, respectively,
W_{fc}, W_{oc}` are diagonal weight matrices for peephole connections. In
our implementation, we use vectors to reprenset these diagonal weight
matrices. The :math:`b` terms denote bias vectors (:math:`b_i` is the input
gate bias vector), :math:`\sigma` is the non-line activations, such as
logistic sigmoid function, and :math:`i, f, o` and :math:`c` are the input
gate, forget gate, output gate, and cell activation vectors, respectively,
all of which have the same size as the cell output activation vector :math:`h`.
The :math:`\odot` is the element-wise product of the vectors. :math:`act_g`
and :math:`act_h` are the cell input and cell output activation functions
and `tanh` is usually used for them. :math:`
\\
tilde{c_t}` is also called
candidate hidden state, which is computed based on the current input and
The :math:`\odot` is the element-wise product of the vectors. :math:`act_g`
and :math:`act_h` are the cell input and cell output activation functions
and `tanh` is usually used for them. :math:`
\\
tilde{c_t}` is also called
candidate hidden state, which is computed based on the current input and
the previous hidden state.
Set `use_peepholes` to `False` to disable peephole connection. The formula
...
...
@@ -251,38 +252,38 @@ def dynamic_lstm(input,
Users can choose to use fully-connect layer before LSTM layer.
Args:
input(Variable): The input of dynamic_lstm layer, which supports
variable-time length input sequence. The underlying
tensor in this Variable is a matrix with shape
(T X 4D), where T is the total time steps in this
input(Variable): The input of dynamic_lstm layer, which supports
variable-time length input sequence. The underlying
tensor in this Variable is a matrix with shape
(T X 4D), where T is the total time steps in this
mini-batch, D is the hidden size.
size(int): 4 * hidden size.
param_attr(ParamAttr): The parameter attribute for the learnable
hidden-hidden weights.
param_attr(ParamAttr): The parameter attribute for the learnable
hidden-hidden weights.
- The shape is (D x 4D), where D is the hidden
size.
- The shape is (D x 4D), where D is the hidden
size.
- Weights = {:math:`W_{ch}, W_{ih},
\
W_{fh}, W_{oh}`}
bias_attr(ParamAttr): The bias attribute for the learnable bias
weights, which contains two parts, input-hidden
bias weights and peephole connections weights if
setting `use_peepholes` to `True`.
weights, which contains two parts, input-hidden
bias weights and peephole connections weights if
setting `use_peepholes` to `True`.
1. `use_peepholes = False`
- The shape is (1 x 4D).
1. `use_peepholes = False`
- The shape is (1 x 4D).
- Biases = {:math:`b_c, b_i, b_f, b_o`}.
2. `use_peepholes = True`
- The shape is (1 x 7D).
2. `use_peepholes = True`
- The shape is (1 x 7D).
- Biases = { :math:`b_c, b_i, b_f, b_o, W_{ic},
\
W_{fc}, W_{oc}`}.
use_peepholes(bool): Whether to enable diagonal/peephole connections,
use_peepholes(bool): Whether to enable diagonal/peephole connections,
default `True`.
is_reverse(bool): Whether to compute reversed LSTM, default `False`.
gate_activation(str): The activation for input gate, forget gate and
output gate. Choices = ["sigmoid", "tanh", "relu",
gate_activation(str): The activation for input gate, forget gate and
output gate. Choices = ["sigmoid", "tanh", "relu",
"identity"], default "sigmoid".
cell_activation(str): The activation for cell output. Choices = ["sigmoid",
cell_activation(str): The activation for cell output. Choices = ["sigmoid",
"tanh", "relu", "identity"], default "tanh".
candidate_activation(str): The activation for candidate hidden state.
Choices = ["sigmoid", "tanh", "relu", "identity"],
...
...
@@ -2027,3 +2028,57 @@ def warpctc(input, label, blank=0, norm_by_times=False, **kwargs):
attrs
=
{
'blank'
:
blank
,
'norm_by_times'
:
norm_by_times
})
return
loss_out
def
sequence_reshape
(
input
,
new_dim
):
"""
**Sequence Reshape Layer**
This layer will rearrange the input sequences. The new dimension is set by
user. Length of each sequence is computed according to original length,
original dimension and new dimension. The following example will help to
illustrate the function of this layer:
.. code-block:: text
x is a LoDTensor:
x.lod = [[0, 2, 6]]
x.data = [[1, 2], [3, 4],
[5, 6], [7, 8], [9, 10], [11, 12]]
x.dims = [6, 2]
set new_dim = 4
then out is a LoDTensor:
out.lod = [[0, 1, 3]]
out.data = [[1, 2, 3, 4],
[5, 6, 7, 8], [9, 10, 11, 12]]
out.dims = [3, 4]
Currently, only 1-level LoDTensor is supported and please make sure
(original length * original dimension) can be divided by new dimension with
no remainder for each sequence.
Args:
input (Variable): (LodTensor, default: LoDTensor<float>), a 2-D LoDTensor
with shape being [N, M] where M for dimension.
new_dim (int): New dimension which the input LoDTensor is reshaped to.
Returns:
Variable: Reshaped LoDTensor according to new dimension.
Examples:
.. code-block:: python
x = fluid.layers.data(name='x', shape=[5, 20],
dtype='float32', lod_level=1)
x_reshaped = layers.sequence_reshape(input=x, new_dim=10)
"""
helper
=
LayerHelper
(
'sequence_reshape'
,
**
locals
())
out
=
helper
.
create_tmp_variable
(
helper
.
input_dtype
())
helper
.
append_op
(
type
=
'sequence_reshape'
,
inputs
=
{
'X'
:
[
input
]},
outputs
=
{
'Out'
:
[
out
]},
attrs
=
{
'new_dim'
:
new_dim
})
return
out
python/paddle/v2/fluid/tests/book_distribute/notest_dist_image_classification.py
0 → 100644
浏览文件 @
680aec21
#Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
#
#Licensed under the Apache License, Version 2.0 (the "License");
#you may not use this file except in compliance with the License.
#You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
#Unless required by applicable law or agreed to in writing, software
#distributed under the License is distributed on an "AS IS" BASIS,
#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#See the License for the specific language governing permissions and
#limitations under the License.
from
__future__
import
print_function
import
sys
import
paddle.v2
as
paddle
import
paddle.v2.fluid
as
fluid
import
os
import
sys
TRAINERS
=
5
BATCH_SIZE
=
128
PASS_NUM
=
100
def
resnet_cifar10
(
input
,
depth
=
32
):
def
conv_bn_layer
(
input
,
ch_out
,
filter_size
,
stride
,
padding
,
act
=
'relu'
):
tmp
=
fluid
.
layers
.
conv2d
(
input
=
input
,
filter_size
=
filter_size
,
num_filters
=
ch_out
,
stride
=
stride
,
padding
=
padding
,
act
=
None
,
bias_attr
=
False
)
return
fluid
.
layers
.
batch_norm
(
input
=
tmp
,
act
=
act
)
def
shortcut
(
input
,
ch_in
,
ch_out
,
stride
):
if
ch_in
!=
ch_out
:
return
conv_bn_layer
(
input
,
ch_out
,
1
,
stride
,
0
,
None
)
else
:
return
input
def
basicblock
(
input
,
ch_in
,
ch_out
,
stride
):
tmp
=
conv_bn_layer
(
input
,
ch_out
,
3
,
stride
,
1
)
tmp
=
conv_bn_layer
(
tmp
,
ch_out
,
3
,
1
,
1
,
act
=
None
)
short
=
shortcut
(
input
,
ch_in
,
ch_out
,
stride
)
return
fluid
.
layers
.
elementwise_add
(
x
=
tmp
,
y
=
short
,
act
=
'relu'
)
def
layer_warp
(
block_func
,
input
,
ch_in
,
ch_out
,
count
,
stride
):
tmp
=
block_func
(
input
,
ch_in
,
ch_out
,
stride
)
for
i
in
range
(
1
,
count
):
tmp
=
block_func
(
tmp
,
ch_out
,
ch_out
,
1
)
return
tmp
assert
(
depth
-
2
)
%
6
==
0
n
=
(
depth
-
2
)
/
6
conv1
=
conv_bn_layer
(
input
=
input
,
ch_out
=
16
,
filter_size
=
3
,
stride
=
1
,
padding
=
1
)
res1
=
layer_warp
(
basicblock
,
conv1
,
16
,
16
,
n
,
1
)
res2
=
layer_warp
(
basicblock
,
res1
,
16
,
32
,
n
,
2
)
res3
=
layer_warp
(
basicblock
,
res2
,
32
,
64
,
n
,
2
)
pool
=
fluid
.
layers
.
pool2d
(
input
=
res3
,
pool_size
=
8
,
pool_type
=
'avg'
,
pool_stride
=
1
)
return
pool
def
vgg16_bn_drop
(
input
):
def
conv_block
(
input
,
num_filter
,
groups
,
dropouts
):
return
fluid
.
nets
.
img_conv_group
(
input
=
input
,
pool_size
=
2
,
pool_stride
=
2
,
conv_num_filter
=
[
num_filter
]
*
groups
,
conv_filter_size
=
3
,
conv_act
=
'relu'
,
conv_with_batchnorm
=
True
,
conv_batchnorm_drop_rate
=
dropouts
,
pool_type
=
'max'
)
conv1
=
conv_block
(
input
,
64
,
2
,
[
0.3
,
0
])
conv2
=
conv_block
(
conv1
,
128
,
2
,
[
0.4
,
0
])
conv3
=
conv_block
(
conv2
,
256
,
3
,
[
0.4
,
0.4
,
0
])
conv4
=
conv_block
(
conv3
,
512
,
3
,
[
0.4
,
0.4
,
0
])
conv5
=
conv_block
(
conv4
,
512
,
3
,
[
0.4
,
0.4
,
0
])
drop
=
fluid
.
layers
.
dropout
(
x
=
conv5
,
dropout_prob
=
0.5
)
fc1
=
fluid
.
layers
.
fc
(
input
=
drop
,
size
=
512
,
act
=
None
)
bn
=
fluid
.
layers
.
batch_norm
(
input
=
fc1
,
act
=
'relu'
)
drop2
=
fluid
.
layers
.
dropout
(
x
=
bn
,
dropout_prob
=
0.5
)
fc2
=
fluid
.
layers
.
fc
(
input
=
drop2
,
size
=
512
,
act
=
None
)
return
fc2
classdim
=
10
data_shape
=
[
3
,
32
,
32
]
images
=
fluid
.
layers
.
data
(
name
=
'pixel'
,
shape
=
data_shape
,
dtype
=
'float32'
)
label
=
fluid
.
layers
.
data
(
name
=
'label'
,
shape
=
[
1
],
dtype
=
'int64'
)
net_type
=
"vgg"
if
len
(
sys
.
argv
)
>=
2
:
net_type
=
sys
.
argv
[
1
]
if
net_type
==
"vgg"
:
print
(
"train vgg net"
)
net
=
vgg16_bn_drop
(
images
)
elif
net_type
==
"resnet"
:
print
(
"train resnet"
)
net
=
resnet_cifar10
(
images
,
32
)
else
:
raise
ValueError
(
"%s network is not supported"
%
net_type
)
predict
=
fluid
.
layers
.
fc
(
input
=
net
,
size
=
classdim
,
act
=
'softmax'
)
cost
=
fluid
.
layers
.
cross_entropy
(
input
=
predict
,
label
=
label
)
avg_cost
=
fluid
.
layers
.
mean
(
x
=
cost
)
optimizer
=
fluid
.
optimizer
.
Adam
(
learning_rate
=
0.001
)
optimize_ops
,
params_grads
=
optimizer
.
minimize
(
avg_cost
)
accuracy
=
fluid
.
evaluator
.
Accuracy
(
input
=
predict
,
label
=
label
)
train_reader
=
paddle
.
batch
(
paddle
.
reader
.
shuffle
(
paddle
.
dataset
.
cifar
.
train10
(),
buf_size
=
128
*
10
),
batch_size
=
BATCH_SIZE
)
place
=
fluid
.
CPUPlace
()
exe
=
fluid
.
Executor
(
place
)
t
=
fluid
.
DistributeTranspiler
()
# all parameter server endpoints list for spliting parameters
pserver_endpoints
=
os
.
getenv
(
"PSERVERS"
)
# server endpoint for current node
current_endpoint
=
os
.
getenv
(
"SERVER_ENDPOINT"
)
# run as trainer or parameter server
training_role
=
os
.
getenv
(
"TRAINING_ROLE"
,
"TRAINER"
)
# get the training role: trainer/pserver
t
.
transpile
(
optimize_ops
,
params_grads
,
pservers
=
pserver_endpoints
,
trainers
=
TRAINERS
)
if
training_role
==
"PSERVER"
:
if
not
current_endpoint
:
print
(
"need env SERVER_ENDPOINT"
)
exit
(
1
)
print
(
"start pserver at:"
,
current_endpoint
)
pserver_prog
=
t
.
get_pserver_program
(
current_endpoint
)
pserver_startup
=
t
.
get_startup_program
(
current_endpoint
,
pserver_prog
)
exe
.
run
(
pserver_startup
)
exe
.
run
(
pserver_prog
)
print
(
"pserver run end"
)
elif
training_role
==
"TRAINER"
:
print
(
"start trainer"
)
trainer_prog
=
t
.
get_trainer_program
()
feeder
=
fluid
.
DataFeeder
(
place
=
place
,
feed_list
=
[
images
,
label
])
exe
.
run
(
fluid
.
default_startup_program
())
for
pass_id
in
range
(
PASS_NUM
):
accuracy
.
reset
(
exe
)
for
data
in
train_reader
():
loss
,
acc
=
exe
.
run
(
trainer_prog
,
feed
=
feeder
.
feed
(
data
),
fetch_list
=
[
avg_cost
]
+
accuracy
.
metrics
)
pass_acc
=
accuracy
.
eval
(
exe
)
print
(
"loss:"
+
str
(
loss
)
+
" acc:"
+
str
(
acc
)
+
" pass_acc:"
+
str
(
pass_acc
))
# this model is slow, so if we can train two mini batch, we think it works properly.
print
(
"trainer run end"
)
else
:
print
(
"environment var TRAINER_ROLE should be TRAINER os PSERVER"
)
exit
(
1
)
python/paddle/v2/fluid/tests/test_layers.py
浏览文件 @
680aec21
...
...
@@ -216,6 +216,14 @@ class TestBook(unittest.TestCase):
self
.
assertIsNotNone
(
x
)
print
(
str
(
program
))
def
test_sequence_reshape
(
self
):
program
=
Program
()
with
program_guard
(
program
):
x
=
layers
.
data
(
name
=
'x'
,
shape
=
[
8
],
dtype
=
'float32'
,
lod_level
=
1
)
out
=
layers
.
sequence_reshape
(
input
=
x
,
new_dim
=
16
)
self
.
assertIsNotNone
(
out
)
print
(
str
(
program
))
if
__name__
==
'__main__'
:
unittest
.
main
()
python/paddle/v2/fluid/tests/test_sequence_reshape.py
0 → 100644
浏览文件 @
680aec21
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
#
#Licensed under the Apache License, Version 2.0 (the "License");
#you may not use this file except in compliance with the License.
#You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
#Unless required by applicable law or agreed to in writing, software
#distributed under the License is distributed on an "AS IS" BASIS,
#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#See the License for the specific language governing permissions and
#limitations under the License.
import
unittest
import
numpy
as
np
import
math
from
op_test
import
OpTest
class
TestSequenceReshape
(
OpTest
):
def
setUp
(
self
):
self
.
op_type
=
'sequence_reshape'
dimension
=
12
x_lod
=
[[
0
,
4
,
5
,
8
,
11
]]
x
=
np
.
random
.
uniform
(
0.1
,
1
,
[
11
,
24
]).
astype
(
'float32'
)
self
.
inputs
=
{
'X'
:
(
x
,
x_lod
)}
self
.
attrs
=
{
'new_dim'
:
dimension
}
out
,
out_lod
=
self
.
compute_output
(
x
,
x_lod
,
dimension
)
self
.
outputs
=
{
'Out'
:
(
out
,
out_lod
)}
def
compute_output
(
self
,
x
,
x_lod
,
dimension
):
x_width
=
x
.
shape
[
1
]
out_lod
=
[[
0
]]
for
i
in
xrange
(
len
(
x_lod
[
0
])
-
1
):
seq_len
=
x_lod
[
0
][
i
+
1
]
-
x_lod
[
0
][
i
]
offset
=
(
seq_len
*
x_width
)
/
dimension
assert
int
(
offset
)
*
dimension
==
seq_len
*
x_width
out_lod
[
0
].
append
(
out_lod
[
0
][
-
1
]
+
int
(
offset
))
out
=
np
.
zeros
(
shape
=
(
out_lod
[
0
][
-
1
],
dimension
)).
astype
(
'float32'
)
out
.
ravel
()[:]
=
x
.
ravel
()[:]
return
out
,
out_lod
def
test_check_output
(
self
):
self
.
check_output
()
def
test_check_grad
(
self
):
self
.
check_grad
([
"X"
],
"Out"
)
class
TestSequenceReshape_reduce
(
TestSequenceReshape
):
def
setUp
(
self
):
self
.
op_type
=
'sequence_reshape'
dimension
=
24
x_lod
=
[[
0
,
4
,
6
,
8
,
12
]]
x
=
np
.
random
.
uniform
(
0.1
,
1
,
[
12
,
12
]).
astype
(
'float32'
)
self
.
inputs
=
{
'X'
:
(
x
,
x_lod
)}
self
.
attrs
=
{
'new_dim'
:
dimension
}
out
,
out_lod
=
self
.
compute_output
(
x
,
x_lod
,
dimension
)
self
.
outputs
=
{
'Out'
:
(
out
,
out_lod
)}
class
TestSequenceReshape_same
(
TestSequenceReshape
):
def
setUp
(
self
):
self
.
op_type
=
'sequence_reshape'
dimension
=
12
x_lod
=
[[
0
,
4
,
6
,
8
,
12
]]
x
=
np
.
random
.
uniform
(
0.1
,
1
,
[
12
,
12
]).
astype
(
'float32'
)
self
.
inputs
=
{
'X'
:
(
x
,
x_lod
)}
self
.
attrs
=
{
'new_dim'
:
dimension
}
out
,
out_lod
=
self
.
compute_output
(
x
,
x_lod
,
dimension
)
self
.
outputs
=
{
'Out'
:
(
out
,
out_lod
)}
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