Sweep `sciplex_hparam` - Githubissues

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

88 stars 23 forks source link

Sweep `sciplex_hparam` #87

Closed MxMstrmn closed 2 years ago

MxMstrmn commented 2 years ago

Resolves #86.

Preliminary sweep for further experiments #84 and #85.

Add a yaml file for the fine-tuning sweep on the sciplex dataset.

hparams that are not swept over come from lincs_rdkit_hparam
hparams that are fixed per drug embedder come from lincs_all_embeddings_hparam
[x] Runs started
[x] Runs finished

review-notebook-app[bot] commented 2 years ago

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

siboehm commented 2 years ago

As discussed we should add:

Evaluation of the variance (the variance scores actually look good here, whereas they were horrible on LINCS)
Sweep the adversarial penalty for lower levels (eg 0.001 - 1.0). I also added this as a task in #86

siboehm commented 2 years ago

There's two configs here now, experiments/sciplex_hparam/config_sciplex_rdkit_hparam.yaml and experiments/sciplex_hparam/config_sciplex_finetuning_hparam.yaml and I'm confused what does what.

MxMstrmn commented 2 years ago

@siboehm, this PR is required for the other experiments and the finetuning_num_genes and finetuning_OOD_prediction. I would propose to merge this even though we have not found the ideal setting (due to metric and/or split).