search

theislab / chemCPA

Code for "Predicting Cellular Responses to Novel Drug Perturbations at a Single-Cell Resolution", NeurIPS 2022.
https://arxiv.org/abs/2204.13545
MIT License
104 stars 24 forks source link

Reproducibility issues #121

Closed regloeb closed 1 year ago

regloeb commented 1 year ago

Congratulations for the great paper and thank you for the amazing work.

I am trying to reproduce some of the experiments and I am really struggling.

Is there a way to run experiments without slurm ? I am trying to use the script chemCPA/manual_seml_sweep.pywith the config file 'config_lincs_rdkit_hparam_sweep.yamland I am getting several bugs. I am using the datasets prepared by the notebooks in thepreprocessing` directory and there seems to be some inconsistency with what the experiment above is expecting.

Some guidance would be greatly appreciated.

Kind regards

jasperhyp commented 1 year ago

Hi @regloeb were you successful? I've been trying to train the model without seml and that proves to be hard. Would appreciate some pointers to essentially make the model training w/o seml!

Edit: Oh I just realized you were referring to slurm instead of seml. Either way, would appreciate some insights if you were successful in that regard.

regloeb commented 1 year ago

Hi @jasperhyp No I was not able to complete the pretraining phase, I tried to modify the code in several places but never got to the end of the process.