Gradient boosting adds up many weak models. Start from a flat guess; each round, look at the residuals (what the model still gets wrong), fit a tiny tree to them, and add a shrunk copy of it. Each tree fixes part of what the previous ones missed.
Gradient boosting adds up many weak models. Start from a flat guess; each round, look at the residuals (what the model still gets wrong), fit a tiny tree to them, and add a shrunk copy of it. Each tree fixes part of what the previous ones missed.