Number classes in JUMP-CP

insitro / ChannelViT

Channel Vision Transformers: An Image Is Worth C x 16 x 16 Words

https://arxiv.org/abs/2309.16108

Other

47 stars 6 forks source link

Number classes in JUMP-CP #5

Closed chaudatascience closed 5 months ago

chaudatascience commented 8 months ago

In the paper, it mentions that JUM-CP contains 160 classes, but in the scripts, num_classes is set to 161.

    python channelvit/main/main_supervised.py \
    ...
    meta_arch.num_classes=161\

I was wondering about the reason behind this difference. Could you please clarify it?

Thank you! Best, Chau

srinivasans-insitro commented 5 months ago

@chaudatascience Sorry for the confusion. The JUMP-CP dataset contains 160 perturbations and 1 control treatment. So, we use num_classes=161 to classify between all conditions (perturbations and controls)

chaudatascience commented 5 months ago

Thank you for the clarification!