binning as preprocessing

bins_from_random_forest.py

import sys
import pandas as pd
from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
from sklearn.datasets import load_iris


def extract_bins(tree):
    '''Extract bins from tree
    '''
    edges = {}
    for feature, threshold in zip(tree.feature, tree.threshold):
        if feature == -2:  # undefined
            continue
        if feature not in edges:
            edges[feature] = []
        edges[feature].append(threshold)

    bins = {}
    for f, e in edges.items():
        print('feature: {}, edges: {}'.format(f, e))
        dummy = e
        dummy.sort()
        dummy.insert(0, sys.float_info.min)
        dummy.append(sys.float_info.max)
        bins[f] = dummy

    return bins


# load some dataset
data = load_iris()

# fit a tree model - rf
clf = RandomForestClassifier(max_depth=2, random_state=0)
clf.fit(data.data, data.target)

# TODO: replace with loop here
dt = clf.estimators_[0]
tree = dt.__dict__['tree_']
bins = extract_bins(tree)

# TODO: use bins for model; implement as preprocessing step