def test_dirty(gaussian_data, fit_intercept, normalize, alpha, beta, positive):
X, y = gaussian_data
n_samples = y.shape[1]
Xty = np.array([xx.T.dot(yy) for xx, yy in zip(X, y)])
alpha_max = np.linalg.norm(Xty, axis=0).max()
beta_max = abs(Xty).max()
alpha *= alpha_max / n_samples
beta *= beta_max / n_samples
est = DirtyModel(alpha=alpha,
beta=beta,
fit_intercept=fit_intercept,
normalize=normalize,
positive=positive)
est.fit(X, y)
assert hasattr(est, 'is_fitted_')
if est.alpha <= est.beta:
assert_allclose(est.coef_specific_, 0.)
elif est.alpha > len(X)**0.5 * est.beta:
assert_allclose(est.coef_shared_, 0.)
if positive:
assert est.coef_.min() >= 0.
var_y = np.var(y, axis=1)
r2 = 1 - (est.residuals_**2).mean(1) / var_y
scores = est.score(X, y)
assert_allclose(r2, scores)
def test_zero_alpha(gaussian_data):
X, y = gaussian_data
est0 = IndRewLasso(alpha=len(X) * [0.])
est1 = IndLasso(alpha=len(X) * [0.])
est2 = GroupLasso(alpha=0.)
est3 = DirtyModel(alpha=0., beta=0.)
est4 = MultiLevelLasso(alpha=0.)
est5 = ReMTW(alpha=0., beta=0.)
est6 = MTW(alpha=0., beta=0.)
models = [est0, est1, est2, est3, est4, est5, est6]
for model in models:
with pytest.warns(UserWarning, match='LinearRegression'):
model.fit(X, y)
def test_slow_convergence(gaussian_data):
X, y = gaussian_data
max_iter = 1
est0 = IndRewLasso(alpha=len(X) * [0.01], max_iter=max_iter)
est1 = IndLasso(alpha=len(X) * [0.01], max_iter=max_iter)
est2 = GroupLasso(alpha=0.01, max_iter=max_iter)
est3 = DirtyModel(alpha=0.01, beta=0.01, max_iter=max_iter)
est4 = MultiLevelLasso(alpha=0.01, max_iter=max_iter)
est5 = ReMTW(alpha=0.01, beta=0.01, max_iter=max_iter)
est6 = MTW(alpha=0.01, beta=0.01, max_iter=max_iter)
models = [est0, est1, est2, est3, est4, est5, est6]
for model in models:
with pytest.warns(ConvergenceWarning, match='number of iterations'):
model.fit(X, y)
xty = np.array([xx.T.dot(yy) for xx, yy in zip(X, y)])
beta_max = abs(xty).max() / n_samples
alpha_max = np.linalg.norm(xty, axis=0).max() / n_samples
alphas = np.linspace(0.02, 1.1, 40) * alpha_max
betas = np.linspace(0.02, 1.1, 40) * beta_max
alphas_mesh, betas_mesh = np.meshgrid(alphas, alphas)
alphas_mesh = alphas_mesh.flatten()
betas_mesh = betas_mesh.flatten()
###########################################################
# For each (alpha, beta) we check if W_1 = 0 or W_2 = 0
type_of_model = []
for alpha, beta in zip(alphas_mesh, betas_mesh):
dirty = DirtyModel(alpha=alpha, beta=beta)
dirty.fit(X, y)
W1 = abs(dirty.coef_shared_).max()
W2 = abs(dirty.coef_specific_).max()
if W1 and not W2:
type_of_model.append(0)
elif W2 and not W1:
type_of_model.append(1)
elif W1 and W2:
type_of_model.append(2)
else:
type_of_model.append(3)
type_of_model = np.array(type_of_model)
###################################################
# Plot nature of model depending on hyperparameters
def test_warmstart(gaussian_data):
X, y = gaussian_data
n_samples = y.shape[1]
Xty = np.array([xx.T.dot(yy) for xx, yy in zip(X, y)])
alpha_max = np.linalg.norm(Xty, axis=0).max()
alpha1 = 0.6 * alpha_max / n_samples
alpha2 = 0.05 * alpha_max / n_samples
est = DirtyModel(alpha=alpha1, beta=alpha1, warm_start=True, tol=1e-5)
est.fit(X, y)
est.alpha = alpha2
est.fit(X, y)
assert hasattr(est, 'is_fitted_')
assert_allclose(est.coef_specific_, 0.)
coef1 = est.coef_.copy()
est = DirtyModel(alpha=alpha2, beta=alpha1, tol=1e-5)
est.fit(X, y)
coef2 = est.coef_
assert_allclose(coef1, coef2, 1e-4)