Fix issue where early stopping rounds wasn't actually stopping training.
0.2.21
Added the ability to force children weights to contain the parent weight, also fixed bug where learning rate wasn't applied to the root node.
0.2.20
Added logging output functionality to forust.
0.2.19
Fixed bug where right node was not always constrained by upper and lower bounds in the `MissingImputer` splitter.
0.2.18
This release adds a new contribution calculation method, that looks at the change of a records probability or being the positive class. Also adds a missing node treatment method that calculates the missing node as the average of the left and right nodes weight.
0.2.17
This release introduces a `missing_node_treatment` parameter to adjust how missing values are treated when they are split out into their own node.