Have you tried to adapt DANCE method for Tabula Muris dataset?

[XiHuYan]

Thanks for sharing this excellent work!

I read you paper and noticed that experiments results didn't show the performance of DANCE method (Saito et al., 2020) on the Tabula Muris dataset. I understand that the backbone in DANCE is a pretrained ResNET while MLP is used for TM dataset. Have you ever tried to adapt the architecture of DANCE for the TM dataset, or is there some reasons (intuitions) that DANCE cannot be adapted for TM dataset?

Thanks again :)

[saurabhgarg1996]

Thanks for your question.

We didn't try to adapt the architecture of DANCE for the TM dataset. I can try and get back to you in a few more days. We would need to just add its support here:

Open-Set-Label-Shift/src/baseline_utils/danceSaito20_model_utils.py

Line 27 in f4424da

def get_model_dance(arch, dataset, num_classes, pretrained, learning_rate, weight_decay, features = False, temp_scale=False, pretrained_model_dir=None):

.