Cross-Validation Playground
Watch validation folds move through the dataset and compare both average performance and variability.
What to observe while you experiment
Cross-validation estimates how a modelling workflow behaves across multiple train/validation partitions. The fold geometry matters: ordinary K-Fold, stratification, groups and time-aware splitting encode different assumptions about which observations may be separated.
Experiment deliberately
Run all folds, then change the split method while keeping the dataset/model fixed. Predict how class balance or dependency leakage should change.
Validation is a sampling procedure. Use the fold visualiser to inspect which observations are training vs validation in each iteration, then compare fold-to-fold variability.
Ready.
Fold map
Reset prepares the folds. Run next fold generates one model/result at a time; completed folds become available for inspection.