Skip to content

evalgate 0.1.0

Choose a tag to compare

View all tags
@delcenjo delcenjo released this 29 Jun 19:35
v0.1.0
b9f1963

First working release.

  • proportions: two-proportion z-test on two aggregate accuracies and their sample sizes.
  • paired: McNemar's test on paired per-example correctness, exact for small samples and chi-squared for large ones.
  • A four-way verdict (improvement, unchanged, noise, regression) that fails CI only on a significant regression, with a configurable --alpha.
Assets 2
Loading