Run summary

Learning progress by stage

Train and test error at each stage. The line shows the combined-product error; points show per-product error within that stage; violins show the distribution.

Column analysis

Per-feature statistics for the selected run: split counts, average gain, average samples affected, and average cells added.

Convergence

Per-stage view for the selected run: total and average gain, split types per stage, and model complexity (number of cells).