The new SplitRecord

split_record.pyx

cdef struct SplitRecord:
    # Data to track sample splitting process
    # This structure also store the best split found so far
    SIZE_t feature         # Which feature to split on.
    SIZE_t start
    SIZE_t end
    SIZE_t pos             # Split samples array at the given position,
                           # i.e. count of samples below threshold for feature.
                           # pos is >= end if the node is a leaf.

    double impurity
    double threshold       # Threshold to split at.
    double proxy_improvement  # Proxy for impurity improvement to speed up
                              # computation times
    double improvement     # Impurity improvement given parent node.

    # Use these to compare the current split stats with the best so far
    SIZE_t best_feature
    SIZE_t best_pos
    double best_threshold
    double best_proxy_improvement
    # This will be updated only finally to save some computations
    double best_improvement

    # stats for left partition
    SIZE_t n_left
    double weighted_n_left

    # stats for right partition
    SIZE_t n_right
    double weighted_n_right