Python implementation of stacked generalization classifier, as described here.
Plays nice with sklearn classifiers, or any model classes that have both .fit and .predict methods.
Currently the package is not on PyPi, but is easy to install directly from github via pip using the following command.
pip install -e 'git+http://github.com/dustinstansbury/stacked_generalization.git#egg=stacked_generalization'
The following example builds a stacked generalizer model to classify the digits dataset available in scikits-learn. The three base models (two RandomForest classifiers with different optimization criterion, and a ExtraTreesClassifier) are estimated with 5-fold cross-validation. The outputs of the fit base models are used as features inputs to the LogisticRegression blending model, which is also trained with 5-fold cross-validation. The models are trained on 80 percent of the digits dataset and accuracy is evaluated on the remaining 20 percent.
from sklearn.datasets import load_digits
from stacked_generalizer import StackedGeneralizer
from sklearn.ensemble import RandomForestClassifier, ExtraTreesClassifier
from sklearn.linear_model import LogisticRegression
import numpy as np
VERBOSE = True
N_FOLDS = 5
# load data and shuffle observations
data = load_digits()
X = data.data
y = data.target
shuffle_idx = np.random.permutation(y.size)
X = X[shuffle_idx]
y = y[shuffle_idx]
# hold out 20 percent of data for testing accuracy
train_prct = 0.8
n_train = int(round(X.shape[0]*train_prct))
# define base models
base_models = [RandomForestClassifier(n_estimators=100, n_jobs=-1, criterion='gini'),
RandomForestClassifier(n_estimators=100, n_jobs=-1, criterion='entropy'),
ExtraTreesClassifier(n_estimators=100, n_jobs=-1, criterion='gini')]
# define blending model
blending_model = LogisticRegression()
# initialize multi-stage model
sg = StackedGeneralizer(base_models, blending_model,
n_folds=N_FOLDS, verbose=VERBOSE)
# fit model
sg.fit(X[:n_train],y[:n_train])
# test accuracy
pred = sg.predict(X[n_train:])
pred_classes = [np.argmax(p) for p in pred]
_ = sg.evaluate(y[n_train:], pred_classes)
precision recall f1-score support
0 1.00 1.00 1.00 31
1 0.95 1.00 0.97 39
2 1.00 1.00 1.00 40
3 1.00 0.97 0.99 38
4 1.00 0.97 0.99 37
5 1.00 0.97 0.99 35
6 1.00 1.00 1.00 32
7 0.95 1.00 0.97 37
8 1.00 0.94 0.97 35
9 0.92 0.94 0.93 35
avg / total 0.98 0.98 0.98 359
Confusion Matrix:
[[31 0 0 0 0 0 0 0 0 0]
[ 0 39 0 0 0 0 0 0 0 0]
[ 0 0 40 0 0 0 0 0 0 0]
[ 0 0 0 37 0 0 0 1 0 0]
[ 0 0 0 0 36 0 0 0 0 1]
[ 0 0 0 0 0 34 0 0 0 1]
[ 0 0 0 0 0 0 32 0 0 0]
[ 0 0 0 0 0 0 0 37 0 0]
[ 0 1 0 0 0 0 0 0 33 1]
[ 0 1 0 0 0 0 0 1 0 33]]