A gradient-boosting model is trained with learning_rate = 0.05 and a validation set. The validation loss falls for the first 420 trees, flattens, and starts rising after tree 600. Training loss keeps falling throughout.
What should the analyst do?
Why did you pick that answer? Two or three sentences. The act of articulating it is what builds the judgment — not the click that follows.
Boosting fits each new tree to the residuals of the ensemble so far. After the real signal is captured, the residuals are mostly noise, and further trees learn that noise — training loss keeps improving while validation loss turns up. Early stopping (built into HistGradientBoostingClassifier, LightGBM, and XGBoost) picks the iteration where validation loss bottomed out. Learning rate and tree count trade off: lower rate, more trees, smoother fit.
Zero training loss is memorization. The validation curve is already telling you the model stopped generalizing.
A learning rate of 1.0 makes each tree overshoot; the model would overfit faster and less smoothly, not less.
The validation set is what let you see the problem. Without it you would ship a 1,000-tree model that is worse than a 500-tree one.