This creates a separate _tree module for the low-level Tree class.
Here we add some post-fit methods which add detail to the serialized forest grf_forest_. These additional elements are added to enable some properties in Tree, particularly the value property. To calculate this we store leaf sample weights corresponding to leaf_samples, and leaf values of y also corresponding to leaf_samples.
In addition to value we also add the following properties also found in sklearn
node_count
capacity
n_outputs (always 1 for now)
n_classes
We also modify feature and threshold to return -2 for leaf nodes as sklearn does. We also modify the default n_classes_ fit attribute to 1 rather than None in the case of no cluster data.
This creates a separate
_treemodule for the low-levelTreeclass.Here we add some post-fit methods which add detail to the serialized forest
grf_forest_. These additional elements are added to enable some properties inTree, particularly thevalueproperty. To calculate this we store leaf sample weights corresponding toleaf_samples, and leaf values ofyalso corresponding toleaf_samples.In addition to
valuewe also add the following properties also found in sklearnWe also modify
featureandthresholdto return-2for leaf nodes as sklearn does. We also modify the defaultn_classes_fit attribute to 1 rather thanNonein the case of no cluster data.