Pretraining and running model

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

Pretraining and running model #162

Open xianglin226 opened 8 months ago

xianglin226 commented 8 months ago

Amazing model. Could you provide the codes to preprocess custom data and use it to train a model? I cannot use the provided scripts for running a model even with the indicated datasets and weights. For example, too many variables are needed from the model that are not in the provided h5ad files. Could you provide more instructions? Thanks

MxMstrmn commented 8 months ago

Hi @xianglin226,

Did you try running the manual_seml_sweep.py - this should work for you. Also, preparing the data and running the model are two different steps, could you clarify where you issue lies?

Thanks a lot :)

xianglin226 commented 8 months ago

Thank you for your reply. 'manual_seml_sweep.py' doesn't work for me directly. Could you indicate the required files (such as data and weight files) that are needed for running manual_seml_sweep.py? Besides, during training, the reconstruction losses were negative numbers. Is this correct or something went wrong? Thanks!