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
曾经的那一瞬间
Models
提交
fa0b75f1
M
Models
项目概览
曾经的那一瞬间
/
Models
大约 1 年 前同步成功
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
DevOps
流水线
流水线任务
计划
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
M
Models
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
DevOps
DevOps
流水线
流水线任务
计划
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
流水线任务
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
提交
fa0b75f1
编写于
7月 24, 2020
作者:
C
Chen Chen
提交者:
A. Unique TensorFlower
7月 24, 2020
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Change the dataset_transform_fn argument in InputReader's constructor to transform_and_batch_fn.
PiperOrigin-RevId: 323013252
上级
4e6b28ad
变更
1
隐藏空白更改
内联
并排
Showing
1 changed file
with
17 addition
and
13 deletion
+17
-13
official/core/input_reader.py
official/core/input_reader.py
+17
-13
未找到文件。
official/core/input_reader.py
浏览文件 @
fa0b75f1
...
...
@@ -32,8 +32,9 @@ class InputReader:
dataset_fn
=
tf
.
data
.
TFRecordDataset
,
decoder_fn
:
Optional
[
Callable
[...,
Any
]]
=
None
,
parser_fn
:
Optional
[
Callable
[...,
Any
]]
=
None
,
dataset_transform_fn
:
Optional
[
Callable
[[
tf
.
data
.
Dataset
],
tf
.
data
.
Dataset
]]
=
None
,
transform_and_batch_fn
:
Optional
[
Callable
[
[
tf
.
data
.
Dataset
,
Optional
[
tf
.
distribute
.
InputContext
]],
tf
.
data
.
Dataset
]]
=
None
,
postprocess_fn
:
Optional
[
Callable
[...,
Any
]]
=
None
):
"""Initializes an InputReader instance.
...
...
@@ -48,9 +49,12 @@ class InputReader:
parser_fn: An optional `callable` that takes the decoded raw tensors dict
and parse them into a dictionary of tensors that can be consumed by the
model. It will be executed after decoder_fn.
dataset_transform_fn: An optional `callable` that takes a
`tf.data.Dataset` object and returns a `tf.data.Dataset`. It will be
executed after parser_fn.
transform_and_batch_fn: An optional `callable` that takes a
`tf.data.Dataset` object and an optional `tf.distribute.InputContext` as
input, and returns a `tf.data.Dataset` object. It will be
executed after `parser_fn` to transform and batch the dataset; if None,
after `parser_fn` is executed, the dataset will be batched into
per-replica batch size.
postprocess_fn: A optional `callable` that processes batched tensors. It
will be executed after batching.
"""
...
...
@@ -101,7 +105,7 @@ class InputReader:
self
.
_dataset_fn
=
dataset_fn
self
.
_decoder_fn
=
decoder_fn
self
.
_parser_fn
=
parser_fn
self
.
_
dataset_transform_fn
=
dataset_transform
_fn
self
.
_
transform_and_batch_fn
=
transform_and_batch
_fn
self
.
_postprocess_fn
=
postprocess_fn
def
_read_sharded_files
(
...
...
@@ -214,13 +218,13 @@ class InputReader:
dataset
=
maybe_map_fn
(
dataset
,
self
.
_decoder_fn
)
dataset
=
maybe_map_fn
(
dataset
,
self
.
_parser_fn
)
if
self
.
_dataset_transform_fn
is
not
None
:
dataset
=
self
.
_dataset_transform_fn
(
dataset
)
per_replica_batch_size
=
input_context
.
get_per_replica_batch_size
(
self
.
_global_batch_size
)
if
input_context
else
self
.
_global_batch_size
if
self
.
_transform_and_batch_fn
is
not
None
:
dataset
=
self
.
_transform_and_batch_fn
(
dataset
,
input_context
)
else
:
per_replica_batch_size
=
input_context
.
get_per_replica_batch_size
(
self
.
_global_batch_size
)
if
input_context
else
self
.
_global_batch_size
dataset
=
dataset
.
batch
(
per_replica_batch_size
,
drop_remainder
=
self
.
_drop_remainder
)
dataset
=
dataset
.
batch
(
per_replica_batch_size
,
drop_remainder
=
self
.
_drop_remainder
)
dataset
=
maybe_map_fn
(
dataset
,
self
.
_postprocess_fn
)
return
dataset
.
prefetch
(
tf
.
data
.
experimental
.
AUTOTUNE
)
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