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

random_state=0 does not set seed in muon.tl.leiden, either a bug or unclear documentation #154

Open
AKMHamid opened this issue Dec 12, 2024 · 0 comments
Labels
bug Something isn't working

Comments

@AKMHamid
Copy link

Describe the bug
When running the muon.tl.leiden function, defaulting the random_state or setting it to 0 leads to non-deterministic outcome. Setting it non-zero numbers seems to set an actual seed. I'm not sure if this is a bug or it's so by design, but the documentation wasn't very clear about this and I think it would be awesome if you would add a note about that.

To Reproduce
Here is an MRE:

import pandas as pd
import numpy as np
import scanpy as sc
import mudata as md
import muon as mu


sampleDataA = pd.DataFrame(np.random.randint(0, 1000, size=(1000, 10)), columns=list('ABCDEFGHIJ'))
sampleDataB = pd.DataFrame(np.random.randint(0, 1000, size=(1000, 10)), columns=list('KLMNOPQRST'))

metadataDFA = pd.DataFrame(index=sampleDataA.columns.tolist())
metadataDFB = pd.DataFrame(index=sampleDataB.columns.tolist())

adObjA = sc.AnnData(X=sampleDataA.values, obs=None, var=metadataDFA)
adObjB = sc.AnnData(X=sampleDataB.values, obs=None, var=metadataDFB)


sc.pp.neighbors(adObjA)
sc.pp.neighbors(adObjB)

mdObj1 = md.MuData({'A': adObjA, 'B': adObjB})
mdObj2 = mdObj1.copy()

totalIterations = 50
seed = 0
isNotIdentical = 0

for i in range(totalIterations):
    mu.tl.leiden(mdObj1, resolution=1, key_added='leiden', random_state=seed)
    clusters1 = mdObj1.obs['leiden'].copy()

    mu.tl.leiden(mdObj2, resolution=1, key_added='leiden', random_state=seed)
    clusters2 = mdObj2.obs['leiden'].copy()

    if not clusters1.equals(clusters2):
        isNotIdentical += 1

print(f"{isNotIdentical} iterations out of {totalIterations} returned non-identical results")

Expected behaviour
I expected that setting random_state=0 means that the seed=0, i.e. deterministic. In the above MRE, after 50 iterations, I typically get 2-4 instances where the clustering is not identical. I also investigated whether it was simply the cluster naming being out of order, but the actual partitioning was different.

System

  • OS: Ubuntu 24.04.1 LTS
  • Python version 3.10.15
  • Versions of libraries involved: Muon 0.1.7, MuData 0.2.3

Additional context
As I mentioned above, this might not be a bug at all, but I still think it would be nice if you would add a note about this to the documentation. Thanks!

@AKMHamid AKMHamid added the bug Something isn't working label Dec 12, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working
Projects
None yet
Development

No branches or pull requests

1 participant