def Lasso(X, r, lambda_, r_var=None):
"""
Function for finding regression coefficients implementing lasso regression
from sklearn.
Note that in order for this function to work in harmony with the other
functions of the project, an empty array is returned in place of the
coefficient variances. The sklearn function does not provide any
variances.
Arguments:
X (array): design matrix
r (array): response vector
lambda_ (float): tuning parameter
r_var (float, optional): variance of response is calculated if none is
provided, defaults to None
Returns:
beta (array): vector of coefficients
beta_var (array): zeroes array of same shape as beta
"""
if r_var is None:
r_var = np.var(r)
model = sklearn_Lasso(alpha=lambda_,
fit_intercept=False,
max_iter=int(1e6))
model.fit(X, r)
beta = model.coef_
# sklearn Lasso does not provide any var
beta_var = np.zeros(beta.shape)
return beta, beta_var
def test_dropin_lasso(sparse_X):
"""Test that our Lasso class behaves as sklearn's Lasso."""
X, y, _, _ = build_dataset(n_samples=20, n_features=30, sparse_X=sparse_X)
alpha_max = np.linalg.norm(X.T.dot(y), ord=np.inf) / X.shape[0]
alpha = alpha_max / 2.
clf = Lasso(alpha=alpha)
clf.fit(X, y)
clf2 = sklearn_Lasso(alpha=alpha)
clf2.fit(X, y)
np.testing.assert_allclose(clf.coef_, clf2.coef_, rtol=1e-5)
check_estimator(Lasso)
def test_dropin_lasso(sparse_X, fit_intercept):
"""Test that our Lasso class behaves as sklearn's Lasso."""
X, y, _, _ = build_dataset(n_samples=20, n_features=30, sparse_X=sparse_X)
alpha_max = np.linalg.norm(X.T.dot(y), ord=np.inf) / X.shape[0]
alpha = alpha_max / 2.
params = dict(alpha=alpha,
fit_intercept=fit_intercept,
tol=1e-10,
normalize=True)
clf = Lasso(**params)
clf.fit(X, y)
clf2 = sklearn_Lasso(**params)
clf2.fit(X, y)
np.testing.assert_allclose(clf.coef_, clf2.coef_, rtol=1e-5)
if fit_intercept:
np.testing.assert_allclose(clf.intercept_, clf2.intercept_)
check_estimator(Lasso)
def test_Lasso(sparse_X, fit_intercept, positive):
"""Test that our Lasso class behaves as sklearn's Lasso."""
X, y = build_dataset(n_samples=20, n_features=30, sparse_X=sparse_X)
if not positive:
alpha_max = norm(X.T.dot(y), ord=np.inf) / X.shape[0]
else:
alpha_max = X.T.dot(y).max() / X.shape[0]
alpha = alpha_max / 2.
params = dict(alpha=alpha,
fit_intercept=fit_intercept,
tol=1e-10,
positive=positive)
clf = Lasso(**params)
clf.fit(X, y)
clf2 = sklearn_Lasso(**params)
clf2.fit(X, y)
assert_allclose(clf.coef_, clf2.coef_, rtol=1e-5)
if fit_intercept:
assert_allclose(clf.intercept_, clf2.intercept_)