[Python] How does the MAE and MSE criterion affect the way a tree splits in scikit's...

I want to understand what actually happens when a node splits depending on the chosen 'criterion' in the latest scikit version of RandomForestRegressor (RFR).

RandomForestClassifier() has criterion options such as GINI Impurity and entropy to measure the quality of a split. In the GINI case, I know that feature boundaries that provide the max decrease in GINI impurity from one level to the next are chosen.

In RandomForestRegressor() the criterion options are 'MSE' and 'MAE'. But what is this error that is being measured and optimised before splitting?

Continue reading...