Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

results not reproducible even random seed '-S' is set #59

Open
jeffchen2000 opened this issue Dec 8, 2018 · 2 comments
Open

results not reproducible even random seed '-S' is set #59

jeffchen2000 opened this issue Dec 8, 2018 · 2 comments
Assignees

Comments

@jeffchen2000
Copy link

when I run shorah.py, I set '-S 100', but the results are not reproducible, (I downloaded and installed shorah in Nov-2018)

@DrYak
Copy link
Member

DrYak commented Feb 15, 2019

  • How much variations are you seeing ?
  • Which version are you using ?

In older versions, there was a bug causing memory corruption. This is fixed in version v1.1.3.

  • This would have caused completely random-looking results in a few corner cases.

The computation of results is dependent on the math library and is thus implementation-dependent. Running it on a Mac laptop and on a Linux HPC cluster can yield slightly different results (and the diri_sampler step is computed in log-space so its typically affected by this type of rounding implementation differences).

  • This would have cause results which are roughly the same, save for fraction-of-% difference in the scores (and probably different ordering in the CSV file).

We're seeing the same kind of difference after having swapped the math engines between Shorah 1.x and upcoming Shorah 2.0

@DrYak DrYak self-assigned this Feb 15, 2019
@DrYak
Copy link
Member

DrYak commented Apr 1, 2020

We will release previews of ShoRAH 2.x shortly.
Please note that in the newer version, -R is now used for seed and -S is for the sigma parameter when analysing strand bias.

It has a different random engine than ShoRAH 1.x, so even by user the same seed parameters, the result will differ.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants