expand each tree by one level, by attempting to split every leaf.
expand each tree by one level, by attempting to split every leaf.
where to save the new tree
the splitter to use to generate candidate splits for each leaf
the evaluator to use to decide which split to use for each leaf
recursively expand multiple times, writing out the new tree at each step
featureImportance should: shuffle data randomly (group on something random then sort on something random?), then stream through and have each instance pick one feature value at random to pass on to the following instance.
featureImportance should: shuffle data randomly (group on something random then sort on something random?), then stream through and have each instance pick one feature value at random to pass on to the following instance. then group by permuted feature and compare error.
add out of time validation
prune a tree to minimize validation error
prune a tree to minimize validation error
Construct a Map[Int,T] from the trainingData for each tree, and then transform the trees using the prune method.
Update the leaves of the current trees from the training set.
Update the leaves of the current trees from the training set.
The leaves target distributions will be set to the summed distributions of the instances in the training set that would get classified to them. Often used to initialize an empty tree.
produce an error object from the current trees and the validation set