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

EXP: Make drug embedder bigger #68

Closed MxMstrmn closed 2 years ago

MxMstrmn commented 2 years ago

Looking at the results in chemical_CPA/simon/plot_sweep_results.ipynb

We observer that GROVER performs weaker than expected. Hypothesis: We assume that this might be related to the size of the GROVER embedding which has 3400 dimensions Potential experiments: Increase size of drug embedders to cope with the large input dimension.

siboehm commented 2 years ago

This is related to #64, I'm running some sweeps with very large embedders there as well.

siboehm commented 2 years ago

Closing, we'll cover this in #69