42 lines
1.1 KiB
Python
42 lines
1.1 KiB
Python
"""
|
|
=========================================================
|
|
Lasso parameter estimation with path and cross-validation
|
|
=========================================================
|
|
|
|
"""
|
|
|
|
import numpy as np
|
|
|
|
################################################################################
|
|
# generate some sparse data to play with
|
|
|
|
n_samples, n_features = 60, 100
|
|
|
|
np.random.seed(1)
|
|
X = np.random.randn(n_samples, n_features)
|
|
coef = 3*np.random.randn(n_features)
|
|
coef[10:] = 0 # sparsify coef
|
|
y = np.dot(X, coef)
|
|
|
|
# add noise
|
|
y += 0.01 * np.random.normal((n_samples,))
|
|
|
|
# Split data in train set and test set
|
|
X_train, y_train = X[:n_samples/2], y[:n_samples/2]
|
|
X_test, y_test = X[n_samples/2:], y[n_samples/2:]
|
|
|
|
|
|
################################################################################
|
|
# Lasso with path and cross-validation
|
|
from scikits.learn.glm.coordinate_descent import LassoPath
|
|
lasso_path = LassoPath()
|
|
|
|
y_pred = lasso_path.fit(X_train, y_train).predict(X_test)
|
|
|
|
print lasso_path
|
|
|
|
# Compute explained variance on test data
|
|
print "r^2 on test data : %f" % (1 - np.linalg.norm(y_test - y_pred)**2
|
|
/ np.linalg.norm(y_test)**2)
|
|
|