A Forest of Questions
Journey

A Forest of Questions

/journeys/a-forest-of-questions

How machines learn to predict, one yes-or-no question at a time: a single decision tree, then boosting, then random forests, all grown in your browser.

Last partXGBoost roadRandom forests road01One tree02Stopping03Small trees04Gradient05Light06Honesty07Breaking08Tuning
Journey · the map

A Forest of Questions

3 roads · 2 out · 8 coming
A1B1C1D1A2B2C2D2A3B3C3D3The clearingBoostingLightGBM heightsThe random forest01 · One tree02 · Stopping03 · Small treesNot written yet04 · GradientNot written yet05 · LightNot written yet06 · HonestyNot written yet07 · BreakingNot written yet08 · TuningNot written yet03 · BaggingNot written yet05 · XGBoostNot written yet
NNot to scale
Stories2 out · 8 coming
BY ROAD
Shared start
01

The tree that predicts

How can a tree help you guess how many ice creams you'll sell, given the weather?
02

Teaching a tree to stop asking

A decision tree that never stops asking questions ends up memorizing its data. How do we tell it when to stop?
Boosting roadBranches off after part 02
03

Many small trees

Small trees, each one fixing the lastNot written yet
04

Where the gradient comes from

Not written yet
LightGBM roadBranches off after part 04
05

What makes it “Light”

The fast one, taken apartNot written yet
06

Keeping the model honest

Validation, early stopping, and the leaks that fool both.Not written yet
07

Where trees break

Not written yet
08

Tuning

Not written yet
XGBoost roadBranches off after part 04
05

XGBoost

The same idea, one derivative furtherNot written yet
Random forests roadBranches off after part 02
03

Bagging and averaging

Many trees, each asked a different questionNot written yet