In the main take-aways section it’s stated, that decision trees
are sensible to hyperparameter tuning
just to clarify, so i understand the meaning, do we mean sensitive to hyperparameter tuning ? or sensible meaning they it’s relatively easy to tune hyperparameters for decision trees?
Considering either the max_depth or the max_leaf_nodes, basically, there is not a good default value that will lead for sure to a good predictor on different datasets. You will have to tune this hyperparameter to get a good predictor that does not overfit and underfit. And you will need to do this tuning for any dataset since the best value will be different each time.
You will see that an ensemble of trees will be less sensitive to this parameter tuning: fixing the depth will give you a good predictor. So we can provide a sensible default value.
I think sensitive was intended rather than sensible here. I just fixed this.