CODEWITHSUNDEEP
pythonnumpypytorchrandom-seed
user637140
user637140

Reputation: 1101

Why doesn't setting `random.seed(42)` give me identical results in pytorch?

I am setting random seed for both random and numpy.random at the beginning of my main file:

import random
import numpy as np
np.random.seed(42)
random.seed(42)
import torch

Nevertheless, when I create a Net() object with randomly initialized parameters, it gives a completely different result every time:

net=neuralnet.Net()
print ("initialized params: ", net.fc1.weight)

Note that neuralnet.Net() is in a different file, and is a class that extends torch.nn.Module. it is torch.nn.Module that is randomly initializing net.fc1.weight, not my own code.

How is it possible that when I create a Net() object with randomly initialized parameters, it gives a completely different result every time?

Upvotes: 0

Views: 4177

Answers (2)

Woma
Woma

Reputation: 101

try:

import torch
torch.manual_seed(0)

For further information: https://pytorch.org/docs/stable/notes/randomness.html

Upvotes: 3

PythonNoob
PythonNoob

Reputation: 1052

Have you looked at: https://github.com/pytorch/pytorch/issues/7068?

There are some recommendations on how to reproduce the results.

Example:

import sys
import random
import datetime as dt

import numpy as np
import torch

torch.manual_seed(42)
torch.cuda.manual_seed(42)
np.random.seed(42)
random.seed(42)
torch.backends.cudnn.deterministic = True

features = torch.randn(2, 5)

# Print stuff.
fnp = features.view(-1).numpy()

print("Time: {}".format(dt.datetime.now()))
for el in fnp:
    print("{:.20f}".format(el))

print("Python: {}".format(sys.version))
print("Numpy: {}".format(np.__version__))
print("Pytorch: {}".format(torch.__version__))

Upvotes: 0

Related Questions