Skip to content

[Question] Discrepancy in gamma parameter for mmd_poly between paper and implementation #6

Description

@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!

Image

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions