Decision Tree Explorer
Decision trees classify data by asking a sequence of yes/no questions — "is x less than 4.2?" — splitting the plane into axis-aligned regions. Press Step to grow the tree one split at a time and watch accuracy improve.
Tree Visualization
untrained
Class A
Class B
split boundary
misclassified point
Controls
Max depth
Splits shown
0
Leaves
—
Tree depth
—
Train accuracy
—
How it works: At each step the
algorithm tries every possible vertical and horizontal
split and keeps the one that best separates the classes,
measured by Gini impurity — how mixed a
region still is. It repeats inside each region until the
depth limit, a pure region, or too few points. Deeper
trees fit the training data better but can memorize
noise (overfitting) — compare depth 3 vs depth 5 on the
same data.
Splits (in the order the tree adds them)
What's happening here? A decision tree
is like a game of twenty questions. Each split carves
the plane along one axis — first the big separation,
then finer ones inside each region. Points with a ring
around them are misclassified by the current tree; watch
them disappear as you add splits. This greedy "best
split at each step" strategy is exactly what libraries
like scikit-learn implement under the hood.
Premium · From the makers of ReLU.chat
On-Device Chatbot Builder Toolkit
Interactive demos teach one concept at a time. The
On-Device Chatbot Builder Toolkit gives you the full
system: frameworks, templates, and evaluation guides for
building your own privacy-first chatbot that runs
entirely on-device.
Get the Toolkit · $29
