Hello-Justify / randomQA

Random in multiprocessing

When you use numpy.random with torch.utils.data.Dataset, something bad happens.

Prerequisites

• python
• numpy
• torch

Usage

python ori_loader.py

python cur_loader.py 
# equal to 
# python cur_loader.py random

python cur_loader.py numpy

# for train
python cur_loader.py numpy random

# for validation and test, shuffle should be False
python cur_loader.py numpy fix

python cur_loader.py torch