Skip to content

Introducing additional hyperparameter tuning metrics

Choose a tag to compare

@mathias-von-ottenbreit mathias-von-ottenbreit released this 22 Mar 16:51
· 255 commits to main since this release

The following hyperparameter tuning metrics are available:

  • "default". This is the default, using the same methodology as when calculating training loss, like in previous APLR versions.
  • "mse".
  • "mae".
  • "rankability".

Deprecated:

  • group_size_for_validation_group_mse (constructor parameter).
  • get_validation_group_mse (method).