Fork me on GitHub

Lasso on dense and sparse dataΒΆ

We show that linear_model.Lasso provides the same results for dense and sparse data and that in the case of sparse data the speed is improved.

Python source code: lasso_dense_vs_sparse_data.py

print(__doc__)

from time import time
from scipy import sparse
from scipy import linalg

from sklearn.datasets.samples_generator import make_regression
from sklearn.linear_model import Lasso


###############################################################################
# The two Lasso implementations on Dense data
print("--- Dense matrices")

X, y = make_regression(n_samples=200, n_features=5000, random_state=0)
X_sp = sparse.coo_matrix(X)

alpha = 1
sparse_lasso = Lasso(alpha=alpha, fit_intercept=False, max_iter=1000)
dense_lasso = Lasso(alpha=alpha, fit_intercept=False, max_iter=1000)

t0 = time()
sparse_lasso.fit(X_sp, y)
print("Sparse Lasso done in %fs" % (time() - t0))

t0 = time()
dense_lasso.fit(X, y)
print("Dense Lasso done in %fs" % (time() - t0))

print("Distance between coefficients : %s"
      % linalg.norm(sparse_lasso.coef_ - dense_lasso.coef_))

###############################################################################
# The two Lasso implementations on Sparse data
print("--- Sparse matrices")

Xs = X.copy()
Xs[Xs < 2.5] = 0.0
Xs = sparse.coo_matrix(Xs)
Xs = Xs.tocsc()

print("Matrix density : %s %%" % (Xs.nnz / float(X.size) * 100))

alpha = 0.1
sparse_lasso = Lasso(alpha=alpha, fit_intercept=False, max_iter=10000)
dense_lasso = Lasso(alpha=alpha, fit_intercept=False, max_iter=10000)

t0 = time()
sparse_lasso.fit(Xs, y)
print("Sparse Lasso done in %fs" % (time() - t0))

t0 = time()
dense_lasso.fit(Xs.toarray(), y)
print("Dense Lasso done in %fs" % (time() - t0))

print("Distance between coefficients : %s"
      % linalg.norm(sparse_lasso.coef_ - dense_lasso.coef_))
Previous
Next