Unable to reproduce paper's results on MeSH dataset about match HOT 3 CLOSED

I use the "MATCH-NoHierarchy" model (with reg=0) to run.

from match.

Hi, thank you for your interest in our work!

Previously, we tuned model hyperparameters on the MAG dataset. Then, for consistency across datasets, as far as I can recall, we used the same hyperparameters for the MeSH dataset.

In case I misremember, could you try a larger n_probes (e.g., 15)?
https://github.com/yuzhimanhua/MATCH/blob/master/configure/models/MATCH-MeSH.yaml#L6

Or could you try a smaller n_layers (e.g., 2 or even 1)?
https://github.com/yuzhimanhua/MATCH/blob/master/configure/models/MATCH-MeSH.yaml#L5

Thanks a lot!

from match.

Thank you for your reply. After I change to smaller n_layers, it work better.

from match.