forest_patch.py

import errno
import numpy as np
import scipy.io as sio
import time
from sklearn import model_selection
from sklearn.ensemble import RandomForestRegressor
from sklearn.ensemble.base import _partition_estimators
from sklearn.externals.joblib import Parallel, delayed
import types

from regression_datasets import DatasetFactory, dataset_list


def _parallel_helper(obj, methodname, *args, **kwargs):
    """Private helper to workaround Python 2 pickle limitations"""
    return getattr(obj, methodname)(*args, **kwargs)


def forest_regressor_predict(self, X):
    """Copy of the RandomForest Regression predict code, while retaining the entire ensemble in self.all_y_hat """
    # Check data
    X = self._validate_X_predict(X)

    # Assign chunk of trees to jobs
    n_jobs, _, _ = _partition_estimators(self.n_estimators, self.n_jobs)

    # Parallel loop
    all_y_hat = Parallel(n_jobs=n_jobs, verbose=self.verbose,
                         backend="threading")(
        delayed(_parallel_helper)(e, 'predict', X, check_input=False)
        for e in self.estimators_)

    # Reduce
    y_hat = sum(all_y_hat) / len(self.estimators_)

    """ This is the ONLY line changed: Save the decision tree results to the object for external use """
    self.all_y_hat = all_y_hat

    return y_hat


def main():
    for name, func in dataset_list.items():  # a.items():  #
        print(name)
        dataset = func()
        Xtrain, X, ytrain, y = model_selection.train_test_split(
                                        dataset.data,dataset.target,train_size=200,random_state=0)

        # Create the RandomForest regressor, and replace its predict function with
        # a predict that saves the individual regressor outputs to a member variable all_y_hat
        regr = RandomForestRegressor(n_estimators=50)
        regr.predict = types.MethodType(forest_regressor_predict, regr)

        # Fit the model
        regr.fit(Xtrain, ytrain)
        yhat = regr.predict(X)
        Z = np.array(regr.all_y_hat)

        # Save results
        Description = "Random Forest Ensemble (%s)\n%s Generated with %s regressors:" % \
                      (name, regr, regr.n_estimators)

        ################################################################################################################
        sio.savemat('./final/rf/{0}.mat'.format(name), {
            'names': str(regr.estimators_),
            'Z': Z, 'y': y,
            'y_RandomForest': yhat,
            # 'Ztrain': Z_train, # NOTE: Combing DecisionTrees is unsupervised by nature
            #'ytrain': ytrain,
            'samples_per_regressor': 200,
            'regressor_samples_overlap': 200,
            'Ey': np.mean(y),
            'Ey2': np.mean(y ** 2),
            'Description': Description
        })

if __name__ == "__main__":
    start = time.time()
    try:
        main()
        print('Done.')
    except KeyboardInterrupt:
        pass
    except IOError as e:  # catch closing of pipes
        if e.errno != errno.EPIPE:
            raise e
    finally:
        print('Total running time: {0}'.format(time.time() - start))