Max Hutchinson
Max Hutchinson
If the training labels have repeats of label values, then it is increasingly possible that every tree in the ensemble makes the same prediction (even if the input values are...
The maximum value of the Gini impurity is `(n-1)/n`, where `n` is the number of classes. This could cause multitask models to be biased towards modeling multi-class labels more accurately...
Gradient boosted trees can outperform random forests, given proper selection of hyperparameters. In lolo, gradient boosting could be a general component in learner composition, e.g. boosting two linear models against...
Right now, hyperparameter optimization is supported only with the bagged learner, which implements `getLoss` with out of bag estimates. A cross validation learner could give any learner a reasonable `getLoss`...