// interactive walkthrough

How a Decision Tree trains — one split at a time

Pick a dataset and an impurity criterion, then watch the tree grow split by split. Every node shows its impurity; every split shows the information-gain calculation that picked it, with the full table of evaluated candidates ranked by gain. Small enough that you can verify the math by hand.

settings changed — tree has been reset to apply.
01

Task & impurity criterion

// what we measure to pick splits
task
criterion
02

Hyperparameters

// drag to tune · changes reset the tree
03

Training controls

// advance one split, or grow the whole tree
ready
data & current prediction
tree being grown
click ▷ Next split to start.
internal (split) leaf (prediction) last action
04

Why this split?

// impurity reduction math for the highlighted node

Build a split to see the impurity-reduction calculation here.

05

Cost-complexity pruning

// post-hoc: drag ccp_alpha to walk the pruning path

Grow the tree first (or click ⏭ Grow to end), then walk the pruning path with the slider here.

training fit
predictions vs actuals
06

Split history

// one row per split, in growth order

No splits yet.