m53297601
/
Paddle
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from 'e9205c38e8'
${ noResults }
Paddle/python/paddle/fluid/contrib/reader
History
guru4elephant 5ed713d519
remove ctr reader, all functions are satisfied in dataset (#18672) 		6 years ago
..
README.md Add multi process reader (#18115) 6 years ago
__init__.py remove ctr reader, all functions are satisfied in dataset (#18672) 6 years ago
distributed_reader.py Add multi process reader (#18115) 6 years ago

README.md

CTR READER

An multi-thread cpp reader that has the same interface with py_reader. It uses cpp multi-thread to read file and is much more faster then the Python read thread in py_reader.

Currently, it support two types of file:

  • gzip
  • plain text file

and two types of data format:

  • cvs data format is :
    • label dense_fea,dense_fea sparse_fea,sparse_fea
  • the svm data format is :
    • label slot1:fea_sign slot2:fea_sign slot1:fea_sign

Distributed reader

The distributed reader is mainly used by multi-process tasks, and the input must be a batch reader.

Cons:

  • It can be operated conveniently so that different processes can read different data.

Pros:

  • If batch_reader produces training data, and batch_reader loads or preprocesses data for a long time, this data reading method may be slower.