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
机器未来
Paddle
提交
d8c9608f
P
Paddle
项目概览
机器未来
/
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
提交
d8c9608f
编写于
3月 17, 2017
作者:
Y
Yu Yang
提交者:
GitHub
3月 17, 2017
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #1618 from reyoung/feature/speed_up_converter
Speed up dense converter.
上级
e29b003e
8c8b2efd
变更
1
隐藏空白更改
内联
并排
Showing
1 changed file
with
63 addition
and
5 deletion
+63
-5
paddle/py_paddle/dataprovider_converter.py
paddle/py_paddle/dataprovider_converter.py
+63
-5
未找到文件。
paddle/py_paddle/dataprovider_converter.py
浏览文件 @
d8c9608f
...
...
@@ -16,11 +16,25 @@ import paddle.trainer.PyDataProvider2 as dp2
import
collections
import
swig_paddle
import
numpy
import
itertools
__all__
=
[
'DataProviderConverter'
]
class
IScanner
(
object
):
"""
The scanner will scan Python object two passes, then convert it to Paddle's
argument.
In the first pass, `pre_scan` will be invoked by every data instance, and
then invoke `finish_pre_scan` to arguments. And the second pass do the same
thing except the functions changed to `scan`, `finish_scan`.
During the first pass, a scanner may count the shape of input matrix and
allocate memory for this argument. Then fill the data into this argument
in second pass.
"""
def
__init__
(
self
,
input_type
,
pos
):
self
.
input_type
=
input_type
if
not
isinstance
(
self
.
input_type
,
dp2
.
InputType
):
...
...
@@ -36,10 +50,40 @@ class IScanner(object):
self
.
data_in_gpu
=
swig_paddle
.
isUsingGpu
(
)
and
swig_paddle
.
getTrainerCount
()
==
1
def
pre_scan
(
self
,
dat
):
"""
First pass scan method. During this method, the scanner could count the
data number, and get the total memory size this batch would use.
:param dat: The python object.
"""
pass
def
finish_pre_scan
(
self
,
argument
):
"""
Finish first scan pass. Allocate the memory.
:param argument: Output arguments object.
:type argument: swig_paddle.Arguments
:return:
"""
pass
def
scan
(
self
,
dat
):
"""
Second pass scan method. Copy the data to arguments.
:param dat: The python object.
"""
pass
def
finish_scan
(
self
,
argument
):
"""
Finish second pass. Finalize the resources, etc.
:param argument: Output arguments object.
:type argument: swig_paddle.Arguments
"""
pass
...
...
@@ -51,12 +95,19 @@ class DenseScanner(IScanner):
def
__init__
(
self
,
input_type
,
pos
):
IScanner
.
__init__
(
self
,
input_type
,
pos
)
self
.
__mat__
=
None
self
.
__height__
=
0
def
pre_scan
(
self
,
dat
):
self
.
__height__
+=
1
def
finish_pre_scan
(
self
,
argument
):
self
.
__mat__
=
numpy
.
ndarray
(
shape
=
(
self
.
__height__
,
self
.
input_type
.
dim
),
dtype
=
numpy
.
float32
)
self
.
__height__
=
0
def
scan
(
self
,
dat
):
if
self
.
__mat__
is
None
:
self
.
__mat__
=
numpy
.
array
([
dat
],
dtype
=
'float32'
)
else
:
self
.
__mat__
=
numpy
.
append
(
self
.
__mat__
,
[
dat
],
axis
=
0
)
self
.
__mat__
[
self
.
__height__
]
=
dat
self
.
__height__
+=
1
def
finish_scan
(
self
,
argument
):
assert
isinstance
(
argument
,
swig_paddle
.
Arguments
)
...
...
@@ -163,7 +214,14 @@ class DataProviderConverter(object):
]
for
each_sample
in
dat
:
for
each_step
,
scanner
in
zip
(
each_sample
,
scanners
):
for
each_step
,
scanner
in
itertools
.
izip
(
each_sample
,
scanners
):
scanner
.
pre_scan
(
each_step
)
for
scanner
in
scanners
:
scanner
.
finish_pre_scan
(
argument
)
for
each_sample
in
dat
:
for
each_step
,
scanner
in
itertools
.
izip
(
each_sample
,
scanners
):
scanner
.
scan
(
each_step
)
for
scanner
in
scanners
:
...
...
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