Skip to content

CRAN release 0.3.0
Compare
Choose a tag to compare
@mayer79 mayer79 released this 29 Sep 11:50
· 165 commits to main since this release
0.3.0
be3fa4f
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

This is intended to be the last version before 1.0.0.

Visible changes

  • Grid of ice() and partial_dep(): So far, the default grid strategy "uniform" used pretty() to generate the evaluation points. To provide more predictable grid sizes, and to be more in line with other implementations of partial dependence and ICE, we now use seq() to create the uniform grid.
  • h2_pairwise() and h2_threeway() will now also include 0 values. Use zero = FALSE to drop them, see below. The padding with 0 is done at no computational cost, and will affect only up to pairwise_m and threeway_m features.
  • hstats(): The default number of features considered for three-way interactions has been changed from threeway_m = pairwise_m to the more cautious threeway_m = min(pairwise_m, 5L). Furthermore, threeway_m is capped at pairwise_m.
  • The print() method of summary.hstats() is less verbose.

Improvements

  • h2_overall(), h2_pairwise(), h2_threeway(), plot.hstats(), and summary.hstats() have received an argument zero = TRUE. Set to FALSE to drop statistics having value 0.
  • perm_importance() and average_loss() will now recycle a univariate response when combined with multivariate predictions. This is useful, e.g., when the prediction function represents the predictions of multiple models that should be evaluated against a common response.

Bug fixes

  • All progress bars were initialized 1 step too late.
  • perm_importance() and average_loss() would fail for "mlogloss" in case the response y was univariate and non-factor/non-character.

Other changes

  • All available H-statistics are now calculated within hstats() and attached to the resulting object. Each statistic is stored as list with numerator and denominator matrices/vectors. The functions h2(), h2_overall(), h2_pairwise(), and h2_threeway(), print.hstats(), summary().hstats(), plot.hstats() will use these without having to recalculate the required numerators and denominators. The results, however, are unchanged.
Assets 2
Loading