Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

I can't reproduce results

See original GitHub issue

❓ Questions and Help

I’m trying to reproduce results for the same experiment.

I’ve tried everything discussed here but nothing worked.

I added the following lines of code in tools/train_net.py:

+torch.backends.cudnn.benchmark = False
+torch.backends.cudnn.deterministic = True
+
+torch.manual_seed(42)
+torch.cuda.manual_seed(42)
+torch.cuda.manual_seed_all(42)
+random.seed(42)
+np.random.seed(42)

Also added the following in maskrcnn_benchmark/data/build.py

data_loader = torch.utils.data.DataLoader(
            dataset,
            num_workers=num_workers,
            batch_sampler=batch_sampler,
            collate_fn=collator,
+           worker_init_fn=np.random.seed(42),
        )

To see if it works, I train for only180 iterations.

any clue?

Issue Analytics

State:
Created 5 years ago
Comments:9 (4 by maintainers)

Top GitHub Comments

1reaction

wesleylpcommented, Jan 23, 2019

Sure, I will try that! Thanks for your help !

1reaction

wesleylpcommented, Jan 23, 2019

I’ll try that and let you know!

Top Results From Across the Web

What should I do if I cannot reproduce experimental published ...

Easiest solution is to invite them over and let them do the experiment. You will find what they are doing different and then...

Six factors affecting reproducibility in life science research and ...

There are several reasons why an experiment cannot be replicated. ... Many studies claim a significant result, but their findings cannot be reproduced....

Why Can't I Reproduce Their Results?

Perhaps your data isn't clean in the first place? Perhaps it has outliers or invalid entries? Perhaps there is a bug in your...

Why can't we reproduce so many scientific findings?

Being able to do an experiment over again and confirm the results is actually a crucial aspect of science, called reproducibility.

Having hard times reproducing your experiments? - Westburg

Although there is a growing alarm about results that cannot be reproduced, the 'reproducibility crisis' is older than we think and goes back...