Batch size?

[TKassis]

Thank you for this useful package. Looking at the CORN paper, and just some quick tests, do we know why CORN seems to perform slightly better with small batch sizes? I would think larger batch size would perform better especially if dealing with 50+ ordinal classes.

Do we have some intuition with regards to how number of ordinal classes might affect optimal batch size?

[rasbt]

Hi there,

that's a good question. I treated the batch size as a hyperparameter and off the top of my head, I don't have a good explanation here.

[TKassis]

Got it, thanks!