Calculates node_values at fit time rather than at access time. Compared to #55 this adds time to training but keeps the size of the serialized forest small, since we calculate the node values rather than store all the y-values and weights required to build them in Tree.
Calculates node_values at fit time rather than at access time. Compared to #55 this adds time to training but keeps the size of the serialized forest small, since we calculate the node values rather than store all the y-values and weights required to build them in Tree.