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
4b950951
P
Paddle
项目概览
PaddlePaddle
/
Paddle
大约 1 年 前同步成功
通知
2298
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
4b950951
编写于
6月 29, 2018
作者:
F
fengjiayi
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Add unittests and fix a few bugs
上级
ba538012
变更
6
隐藏空白更改
内联
并排
Showing
6 changed file
with
206 addition
and
8 deletion
+206
-8
paddle/fluid/framework/details/data_balance_op_handle.cc
paddle/fluid/framework/details/data_balance_op_handle.cc
+0
-1
paddle/fluid/framework/details/fetch_op_handle.cc
paddle/fluid/framework/details/fetch_op_handle.cc
+1
-1
paddle/fluid/framework/details/multi_devices_graph_builder.cc
...le/fluid/framework/details/multi_devices_graph_builder.cc
+4
-2
paddle/fluid/framework/lod_tensor.cc
paddle/fluid/framework/lod_tensor.cc
+1
-0
paddle/fluid/operators/read_op.cc
paddle/fluid/operators/read_op.cc
+12
-4
python/paddle/fluid/tests/unittests/test_data_balance.py
python/paddle/fluid/tests/unittests/test_data_balance.py
+188
-0
未找到文件。
paddle/fluid/framework/details/data_balance_op_handle.cc
浏览文件 @
4b950951
...
...
@@ -107,7 +107,6 @@ void DataBalanceOpHandle::RunImpl() {
auto
*
tensor_var
=
local_scope
->
FindVar
(
in_var_handles
[
i
]
->
name_
);
PADDLE_ENFORCE
(
tensor_var
->
IsType
<
LoDTensor
>
());
auto
*
tensor
=
tensor_var
->
GetMutable
<
LoDTensor
>
();
PADDLE_ENFORCE
(
places_
[
place_idx
]
==
tensor
->
place
());
lod_tensors
[
data_idx
].
push_back
(
tensor
);
int
ins_size
=
tensor
->
lod
().
empty
()
?
tensor
->
dims
()[
0
]
:
tensor
->
NumElements
();
...
...
paddle/fluid/framework/details/fetch_op_handle.cc
浏览文件 @
4b950951
...
...
@@ -67,8 +67,8 @@ void FetchOpHandle::RunImpl() {
#endif
}
else
{
tensors_
[
i
].
ShareDataWith
(
t
);
tensors_
[
i
].
set_lod
(
t
.
lod
());
}
tensors_
[
i
].
set_lod
(
t
.
lod
());
}
this
->
WaitAndMergeCPUTensors
();
...
...
paddle/fluid/framework/details/multi_devices_graph_builder.cc
浏览文件 @
4b950951
...
...
@@ -216,11 +216,13 @@ std::unique_ptr<SSAGraph> MultiDevSSAGraphBuilder::Build(
}
else
{
// This op runs on all devices, and its output may have parameter's
// gradients.
CreateComputationalOps
(
&
result
,
*
op
,
places_
.
size
());
if
(
op
->
Type
()
==
"read"
)
{
op
->
SetAttr
(
"throw_eof_exp"
,
false
);
CreateComputationalOps
(
&
result
,
*
op
,
places_
.
size
());
const
auto
&
data_var_names
=
op
->
Output
(
"Out"
);
InsertDataBalanceOp
(
&
result
,
data_var_names
);
}
else
{
CreateComputationalOps
(
&
result
,
*
op
,
places_
.
size
());
}
if
(
!
is_forwarding
&&
places_
.
size
()
>
1
)
{
...
...
paddle/fluid/framework/lod_tensor.cc
浏览文件 @
4b950951
...
...
@@ -393,6 +393,7 @@ void LoDTensor::MergeLoDTensor(
new_dim
[
0
]
+=
t
->
dims
()[
0
];
auto
&
lod
=
t
->
lod
();
PADDLE_ENFORCE_EQ
(
new_lod
.
size
(),
lod
.
size
());
for
(
size_t
j
=
0
;
j
<
lod
.
size
();
++
j
)
{
auto
&
sub_lod
=
new_lod
[
j
];
auto
&
offset
=
sub_lod
.
back
();
...
...
paddle/fluid/operators/read_op.cc
浏览文件 @
4b950951
...
...
@@ -67,10 +67,14 @@ class ReadOp : public framework::OperatorBase {
std
::
vector
<
framework
::
LoDTensor
>
ins
;
reader
->
ReadNext
(
&
ins
);
if
(
ins
.
empty
())
{
ins
.
resize
(
out_arg_names
.
size
());
for
(
auto
&
tensor
:
ins
)
{
// data type is not important for subsequent DataBalanceOpHandle
tensor
.
mutable_data
<
float
>
(
framework
::
make_ddim
({
0
}),
dev_place
);
if
(
Attr
<
bool
>
(
"throw_eof_exp"
))
{
PADDLE_THROW
(
"There is no next data."
);
}
else
{
ins
.
resize
(
out_arg_names
.
size
());
for
(
auto
&
tensor
:
ins
)
{
// data type is not important for subsequent DataBalanceOpHandle
tensor
.
mutable_data
<
float
>
(
framework
::
make_ddim
({
0
}),
dev_place
);
}
}
}
PADDLE_ENFORCE_EQ
(
ins
.
size
(),
out_arg_names
.
size
());
...
...
@@ -88,6 +92,10 @@ class ReadOpMaker : public framework::OpProtoAndCheckerMaker {
void
Make
()
override
{
AddInput
(
"Reader"
,
"(ReaderHolder) The executed reader."
);
AddOutput
(
"Out"
,
"(LoDTensor) The output data."
).
AsDuplicable
();
AddAttr
<
bool
>
(
"throw_eof_exp"
,
"If set true, an exception will be thrown when the Reader "
"yields empty (which means there is no next data)."
)
.
SetDefault
(
true
);
AddComment
(
R"DOC(
Read Operator
...
...
python/paddle/fluid/tests/unittests/test_data_balance.py
0 → 100644
浏览文件 @
4b950951
# Copyright (c) 2018 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.
import
unittest
import
paddle.fluid
as
fluid
import
paddle.v2
as
paddle
import
paddle.v2.dataset.mnist
as
mnist
import
numpy
as
np
class
TestDataBalance
(
unittest
.
TestCase
):
def
prepare_data
(
self
):
def
fake_data_generator
():
for
n
in
xrange
(
self
.
total_ins_num
):
yield
np
.
ones
((
3
,
4
))
*
n
,
n
# Prepare data
with
fluid
.
program_guard
(
fluid
.
Program
(),
fluid
.
Program
()):
reader
=
paddle
.
batch
(
fake_data_generator
,
batch_size
=
self
.
batch_size
)
feeder
=
fluid
.
DataFeeder
(
feed_list
=
[
fluid
.
layers
.
data
(
name
=
'image'
,
shape
=
[
3
,
4
],
dtype
=
'float32'
),
fluid
.
layers
.
data
(
name
=
'label'
,
shape
=
[
1
],
dtype
=
'int64'
),
],
place
=
fluid
.
CPUPlace
())
self
.
num_batches
=
fluid
.
recordio_writer
.
convert_reader_to_recordio_file
(
self
.
data_file_name
,
reader
,
feeder
)
def
prepare_lod_data
(
self
):
def
fake_data_generator
():
for
n
in
xrange
(
1
,
self
.
total_ins_num
+
1
):
d1
=
(
np
.
ones
((
n
,
3
))
*
n
).
astype
(
'float32'
)
d2
=
(
np
.
array
(
n
).
reshape
((
1
,
1
))).
astype
(
'int32'
)
yield
d1
,
d2
# Prepare lod data
with
fluid
.
program_guard
(
fluid
.
Program
(),
fluid
.
Program
()):
with
fluid
.
recordio_writer
.
create_recordio_writer
(
filename
=
self
.
lod_data_file_name
)
as
writer
:
eof
=
False
generator
=
fake_data_generator
()
while
(
not
eof
):
data_batch
=
[
np
.
array
([]).
reshape
((
0
,
3
)),
np
.
array
([]).
reshape
(
(
0
,
1
))
]
lod
=
[
0
]
for
_
in
xrange
(
self
.
batch_size
):
try
:
ins
=
generator
.
next
()
except
StopIteration
:
eof
=
True
break
for
i
,
d
in
enumerate
(
ins
):
data_batch
[
i
]
=
np
.
concatenate
(
(
data_batch
[
i
],
d
),
axis
=
0
)
lod
.
append
(
lod
[
-
1
]
+
ins
[
0
].
shape
[
0
])
if
data_batch
[
0
].
shape
[
0
]
>
0
:
for
i
,
d
in
enumerate
(
data_batch
):
t
=
fluid
.
LoDTensor
()
t
.
set
(
data_batch
[
i
],
fluid
.
CPUPlace
())
if
i
==
0
:
t
.
set_lod
([
lod
])
writer
.
append_tensor
(
t
)
writer
.
complete_append_tensor
()
def
setUp
(
self
):
self
.
use_cuda
=
fluid
.
core
.
is_compiled_with_cuda
()
self
.
data_file_name
=
'./data_balance_test.recordio'
self
.
lod_data_file_name
=
'./data_balance_with_lod_test.recordio'
self
.
total_ins_num
=
50
self
.
batch_size
=
10
self
.
prepare_data
()
self
.
prepare_lod_data
()
def
main
(
self
):
main_prog
=
fluid
.
Program
()
startup_prog
=
fluid
.
Program
()
with
fluid
.
program_guard
(
main_prog
,
startup_prog
):
data_reader
=
fluid
.
layers
.
io
.
open_files
(
filenames
=
[
self
.
data_file_name
],
shapes
=
[[
-
1
,
3
,
4
],
[
-
1
,
1
]],
lod_levels
=
[
0
,
0
],
dtypes
=
[
'float32'
,
'int64'
])
if
self
.
use_cuda
:
data_reader
=
fluid
.
layers
.
double_buffer
(
data_reader
)
image
,
label
=
fluid
.
layers
.
read_file
(
data_reader
)
place
=
fluid
.
CUDAPlace
(
0
)
if
self
.
use_cuda
else
fluid
.
CPUPlace
()
exe
=
fluid
.
Executor
(
place
)
exe
.
run
(
startup_prog
)
parallel_exe
=
fluid
.
ParallelExecutor
(
use_cuda
=
self
.
use_cuda
,
main_program
=
main_prog
)
if
(
parallel_exe
.
device_count
>
self
.
batch_size
):
print
(
"WARNING: Unittest TestDataBalance skipped.
\
For the result is not correct when device count
\
is larger than batch size."
)
exit
(
0
)
fetch_list
=
[
image
.
name
,
label
.
name
]
data_appeared
=
[
False
]
*
self
.
total_ins_num
while
(
True
):
try
:
image_val
,
label_val
=
parallel_exe
.
run
(
fetch_list
,
return_numpy
=
True
)
except
fluid
.
core
.
EnforceNotMet
as
ex
:
self
.
assertIn
(
"There is no next data."
,
ex
.
message
)
break
ins_num
=
image_val
.
shape
[
0
]
broadcasted_label
=
np
.
ones
(
(
ins_num
,
3
,
4
))
*
label_val
.
reshape
((
ins_num
,
1
,
1
))
self
.
assertEqual
(
image_val
.
all
(),
broadcasted_label
.
all
())
for
l
in
label_val
:
self
.
assertFalse
(
data_appeared
[
l
[
0
]])
data_appeared
[
l
[
0
]]
=
True
for
i
in
data_appeared
:
self
.
assertTrue
(
i
)
def
main_lod
(
self
):
main_prog
=
fluid
.
Program
()
startup_prog
=
fluid
.
Program
()
with
fluid
.
program_guard
(
main_prog
,
startup_prog
):
data_reader
=
fluid
.
layers
.
io
.
open_files
(
filenames
=
[
self
.
lod_data_file_name
],
shapes
=
[[
-
1
,
3
],
[
-
1
,
1
]],
lod_levels
=
[
1
,
0
],
dtypes
=
[
'float32'
,
'int32'
],
thread_num
=
1
)
ins
,
label
=
fluid
.
layers
.
read_file
(
data_reader
)
place
=
fluid
.
CUDAPlace
(
0
)
if
self
.
use_cuda
else
fluid
.
CPUPlace
()
exe
=
fluid
.
Executor
(
place
)
exe
.
run
(
startup_prog
)
parallel_exe
=
fluid
.
ParallelExecutor
(
use_cuda
=
self
.
use_cuda
,
main_program
=
main_prog
)
if
(
parallel_exe
.
device_count
>
self
.
batch_size
):
print
(
"WARNING: Unittest TestDataBalance skipped.
\
For the result is not correct when device count
\
is larger than batch size."
)
exit
(
0
)
fetch_list
=
[
ins
.
name
,
label
.
name
]
data_appeared
=
[
False
]
*
self
.
total_ins_num
while
(
True
):
try
:
ins_tensor
,
label_tensor
=
parallel_exe
.
run
(
fetch_list
,
return_numpy
=
False
)
except
fluid
.
core
.
EnforceNotMet
as
ex
:
self
.
assertIn
(
"There is no next data."
,
ex
.
message
)
break
ins_val
=
np
.
array
(
ins_tensor
)
label_val
=
np
.
array
(
label_tensor
)
ins_lod
=
ins_tensor
.
lod
()[
0
]
self
.
assertEqual
(
ins_val
.
shape
[
1
],
3
)
self
.
assertEqual
(
label_val
.
shape
[
1
],
1
)
self
.
assertEqual
(
len
(
ins_lod
)
-
1
,
label_val
.
shape
[
0
])
for
i
in
range
(
0
,
len
(
ins_lod
)
-
1
):
ins_elem
=
ins_val
[
ins_lod
[
i
]:
ins_lod
[
i
+
1
]][:]
label_elem
=
label_val
[
i
][
0
]
self
.
assertEqual
(
ins_elem
.
all
(),
label_elem
.
all
())
self
.
assertFalse
(
data_appeared
[
int
(
label_elem
-
1
)])
data_appeared
[
int
(
label_elem
-
1
)]
=
True
for
i
in
data_appeared
:
self
.
assertTrue
(
i
)
def
test_all
(
self
):
self
.
main
()
self
.
main_lod
()
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