search

theislab / scCODA

A Bayesian model for compositional single-cell data analysis
BSD 3-Clause "New" or "Revised" License
147 stars 24 forks source link

Exit code 134 #46

Closed RachellyN closed 2 years ago

RachellyN commented 2 years ago

Hi, I ran scCODA the following way:

counts_per_sample = pd.read_csv("counts2.csv")
counts_adata = dat.from_pandas(counts_per_sample, covariate_columns=['Condition'])
model = mod.CompositionalAnalysis(counts_adata, formula=f"C('Condition', Treatment('ctl'))", reference_cell_type='automatic')
sim_results = model.sample_hmc()
sim_results.summary()

The execution was halted due to Exit code 134. Full output and input files are attached. Thanks, Rachelly.

counts2.csv output2.txt.txt

johannesostner commented 2 years ago

Hi Rachelly, thanks for reporting this error!

This seems to be an issue related to tensorflow. Could you please provide the version numbers of your tensorflow, tensorflwo-probability and scCODA installations? Also, are you trying to run scCODA on a cluster? The error resource temporarily unavailable could also be related to that.

RachellyN commented 2 years ago

Hi johannesostner, Here are my software versions: scCODA: 0.1.4 tensorflow: 2.6.0 tensorflow-probability: 0.14.0

And yes, I am using a cluster.

johannesostner commented 2 years ago

I am unsure whether this is the cause of the error, but the package versions you use are not tested with scCODA. The package is tested against tensorflow 2.4 and tensorflow-probability 2.6. There was an oversight on our side in the 0.1.4 version of scCODA, where the latest versions of both packages were always installed, instead of the recommended ones. Please try downgrading these two packages and upgrading scCODA to 0.1.6.

I hope that this fixes your problem!

RachellyN commented 2 years ago

Thank you again. After upgrading to scCODA 0.016, tensorflow 2.4 and tensorflow-probability 0.12 (there is no version 2.6), this error message is gone!

Zethson commented 2 years ago

@johannesostner pinning Tensorflow to >=2.4 < 2.6 solves the issue, but is rather restrictive for environments which use more tools than sccoda. Would it be possible to identify and fix the root cause?

johannesostner commented 2 years ago

The latest version of scCODA (0.1.7) are compatible with the latest tensorflow (2.8.0) and tensorflow-probability (0.16.0) versions. I could not reproduce the error described here, though. Closing this for now.