test_ridge_crime.py

Makefile

CRIME_DATA_URL := http://archive.ics.uci.edu/ml/machine-learning-databases/communities/communities.data
CRIME_DATA := crime.data

test: $(CRIME_DATA)
	time python test_ridge_crime.py $^
.PHONY: test

$(CRIME_DATA):
	wget -O $@ $(CRIME_DATA_URL)

test_ridge_crime.py

import csv
import sys
import numpy
numpy.random.seed(12345)

def parse_value(x):
    try:
        return float(x)
    except ValueError:
        return numpy.nan

def get_crime_data(lines):
    reader = csv.reader(lines)
    cols = numpy.asarray(list(reader)).T
    cols = cols[5:, :]
    return map(lambda col : map(parse_value, col), cols)

def fill_missing_values(cols):
    cols = numpy.asarray(cols)
    imputed_cols = []
    for i, col in enumerate(cols):
        mask = numpy.isfinite(col)
        fill_value = numpy.median(col[mask])
        imputed_col = numpy.array(col)
        imputed_col[numpy.logical_not(mask)] = fill_value
        imputed_cols.append(imputed_col)
    return numpy.vstack(imputed_cols)

def main():
    if len(sys.argv) != 2:
        print 'usage: data.csv'
        sys.exit(1)
    print 'reading data'
    with open(sys.argv[1]) as f:
        cols = get_crime_data(f)
    print 'filling missing values'
    filled_cols = fill_missing_values(cols)
    data = filled_cols.T

    print 'preparing ridge regression'
    from sklearn.linear_model.ridge import RidgeCV
    from sklearn import cross_validation

    learner = RidgeCV(
        alphas = 2 ** numpy.linspace(-10, 11, 21),
        fit_intercept = True,
        normalize = False
    )
    
    n_examples = data.shape[0]

    # splits = cross_validation.ShuffleSplit(n_examples, n_iterations = 10,
    #    test_fraction = 0.8, indices = True, random_state = 12345)

    splits = [(range(int(n_examples * 0.2)), range(int(n_examples * 0.2), n_examples))]
    for i, (train_indices, test_indices) in enumerate(splits):
        print 'split %d' % i
        train = data[train_indices, :]
        test = data[test_indices, :]
        print '\ttraining examples: %d' % train.shape[0]
        print '\ttesting examples: %d' % test.shape[0]
        x_train = train[:, :-1]
        y_train = train[:, -1]
        print '\tfitting ridge model'
        fit = learner.fit(x_train, y_train)
        x_test = test[:, :-1]
        y_test = test[:, -1]
        print '\t\tbest_alpha = %e' % fit.best_alpha
        print '\tmaking predictions'
        predictions = fit.predict(x_test)
        mse = numpy.mean((predictions - y_test) ** 2)
        print '\ttest mse = %e' % mse

if __name__ == '__main__':
    main()