forest-ranking

#!/usr/bin/env python

import numpy as np
from sklearn.ensemble import ExtraTreesClassifier
import pylab as pl

from mva.categories import (
    Category_Preselection, Category_VBF, Category_Boosted)
from mva.classify import make_dataset, prepare_dataset
from mva.analysis import Analysis
from mva.samples import Higgs
from mva.defaults import TRAIN_FAKES_REGION, TARGET_REGION

analysis = Analysis(
    year=2012,
    systematics=False,
    fakes_region=TRAIN_FAKES_REGION)
analysis.normalize(Category_Preselection)

backgrounds_train = analysis.backgrounds

for category in (Category_VBF, Category_Boosted):
    
    features = category.features
    n_features = len(features)

    signals_train = [
        Higgs(year=2012,
              mass=125,
              modes=category.train_signal_modes,
              trigger=True),
        ]
    
    # create the dataset
    sig, sig_w, bkg, bkg_w = make_dataset(signals_train, backgrounds_train,
                                          fields=features,
                                          category=category,
                                          region=TARGET_REGION)

    sample, labels, weights = prepare_dataset(sig, sig_w, bkg, bkg_w)

    # Build a forest and compute the feature importances
    forest = ExtraTreesClassifier(n_estimators=250, random_state=0)

    forest.fit(sample, labels, sample_weight=weights)
    importances = forest.feature_importances_
    std = np.std([tree.feature_importances_ for tree in forest.estimators_], axis=0)
    indices = np.argsort(importances)[::-1]

    # Print the feature ranking
    print "Feature ranking:"

    for f in range(n_features):
        print "%d. (%f) %s" % (f + 1, importances[indices[f]], features[indices[f]])

    # Plot the feature importances of the forest
    pl.figure()
    pl.title("Feature importances")
    pl.bar(range(n_features), importances[indices],
           color="r", yerr=std[indices], align="center")
    pl.xticks(range(n_features), indices)
    pl.xlim([-1, n_features])
    pl.savefig('ranking_{0}.png'.format(category.name))