Adjust adversarial parameters

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

99 stars 24 forks source link

Closed siboehm closed 2 years ago

siboehm commented 2 years ago

Check the maximal possible accuracy score on LINCS
Adjust adversarial parameters st the accuracy score is worse in the hyperparameter sweeps.

siboehm commented 2 years ago

So in LINCS_full_smiles.h5ad, in the "test" section of the random_split split (which is what is used to calc disentanglement), there are

⇒ Accuracy by always predicting most common drug: 454/12531 = 0.0376

MxMstrmn commented 2 years ago

So i guess this would be the ideal disentanglement score then?

siboehm commented 2 years ago

yes, I think so