Created
November 11, 2011 02:41
-
-
Save agramfort/1357024 to your computer and use it in GitHub Desktop.
parameter scaling by n_samples pb
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import numpy as np | |
from scipy import linalg | |
from sklearn import datasets, svm, linear_model | |
from sklearn.svm import l1_min_c | |
iris = datasets.load_iris() | |
X = iris.data | |
y = iris.target | |
X = X[y != 2] | |
y = y[y != 2] | |
X -= np.mean(X, 0) | |
X2 = np.r_[X, X] | |
y2 = np.r_[y, y] | |
clfs = [linear_model.LogisticRegression(penalty='l1', tol=1e-6), | |
linear_model.LogisticRegression(penalty='l2', tol=1e-6), | |
svm.SVR(kernel='linear', tol=1e-6)] | |
for clf in clfs: | |
for scaling, scaling2 in [(1., 1.), (len(X), len(X2))]: | |
clf.C = 10. / scaling | |
coef_ = clf.fit(X, y).coef_ | |
clf.C = 10. / scaling2 | |
coef2_ = clf.fit(X2, y2).coef_ | |
print "Error : %s" % (linalg.norm(coef2_ - coef_) / linalg.norm(coef_)) | |
clfs = [linear_model.Lasso(tol=1e-6)] | |
for clf in clfs: | |
for scaling, scaling2 in [(1., 1.), (len(X), len(X2))]: | |
clf.alpha = 0.001 * scaling | |
coef_ = clf.fit(X, y).coef_ | |
clf.alpha = 0.001 * scaling2 | |
coef2_ = clf.fit(X2, y2).coef_ | |
print "Error : %s" % (linalg.norm(coef2_ - coef_) / linalg.norm(coef_)) | |
# Output: | |
# Error : 0.13608933675 | |
# Error : 8.56854960126e-16 | |
# Error : 0.124357027952 | |
# Error : 3.02427897402e-16 | |
# Error : 0.000213002674373 | |
# Error : 2.40941736763e-07 | |
# Error : 4.07295351853e-12 | |
# Error : 0.16694490818 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment