[Question] Discrepancy in gamma parameter for mmd_poly between paper and implementation

[ZihaoWang-CV]

Hi authors,

Thank you for your excellent work.
I have a quick question regarding the implementation of the polynomial MMD kernel.

1. In the code, the mmd_poly function appears to use the sklearn default for gamma, which is None (scaled to 1 / n_features).
   videojedi/mmd_polynomial.py#L6
   videojedi/JEDi.py#L19
   

2. In the paper's appendix, it is stated that gamma is set to 1.0.

3. My own test suggests the code with the default gamma=None produces the metric scale reported in the paper (~0.5). When I explicitly set gamma=1.0, the MMD value increases to the order of 1e3.

This leads me to believe the implementation with gamma=None was used for the paper's results. Could you please clarify which is the correct implementation corresponding to the paper's findings?

Thank you for your time and help!