plot_multiestimator_search.py

#!/usr/bin/env python
"""
==============================================================
Grid Search over multiple Pipeline Configurations & Estimators
==============================================================

This example constructs a pipeline that does an optional prescaling step,
then dimensionality reduction followed by prediction with a decision tree
ensemble. It demonstrates the use of GridSearchCV and Pipeline to optimize
over different classes of estimators in a single CV run -- Gradient Boosted
Decision Trees are compared to Random Forests during the grid search.
"""
# Authors: Robert McGibbon, Joel Nothman, Moritz Lotze

from __future__ import print_function, division

import numpy as np
import matplotlib.pyplot as plt
from sklearn.datasets import load_iris
from sklearn.ensemble import GradientBoostingClassifier, RandomForestClassifier
from sklearn.metrics import accuracy_score
from sklearn.model_selection import GridSearchCV, cross_val_predict
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import RobustScaler
from sklearn.decomposition import PCA
from sklearn.feature_selection import SelectKBest, f_classif

print(__doc__)

pipe = Pipeline([
    ('prescale', RobustScaler()),
    ('reduce_dim', PCA()),
    ('classify', RandomForestClassifier())
])

N_FEATURES_OPTIONS = [2, 4, ]
param_grid = [
    {
        'prescale': [RobustScaler(), None],
        'reduce_dim': [PCA(iterated_power=7), ],
        'reduce_dim__n_components': N_FEATURES_OPTIONS,
        'classify': [RandomForestClassifier(), GradientBoostingClassifier()],
    },
    {
        'reduce_dim': [SelectKBest(f_classif)],
        'reduce_dim__k': N_FEATURES_OPTIONS,
        'classify': [RandomForestClassifier(), GradientBoostingClassifier()],
    },
]

grid = GridSearchCV(pipe, cv=3, n_jobs=8, param_grid=param_grid)
iris = load_iris()
grid.fit(iris.data, iris.target)
preds = cross_val_predict(grid, iris.data, iris.target)
accs = accuracy_score(preds, iris.target)
print(accs)