test with large data

Author: mathurinm

gram_test.py

@@ -0,0 +1,57 @@
+# data available at  https://www.dropbox.com/sh/32b3mr3xghi496g/AACNRS_NOsUXU-hrSLixNg0ja?dl=0
+
+
+import time
+from numpy.linalg import norm
+import matplotlib.pyplot as plt
+import numpy as np
+from celer import GroupLasso
+from skglm.gram_solver import group_lasso
+
+X = np.load("design_matrix.npy")
+y = np.load("target.npy")
+groups = np.load("groups.npy")
+weights = np.load("weights.npy")
+# grps = [list(np.where(groups == i)[0]) for i in range(1, 33)]
+
+
+alpha_ratio = 1e-2
+n_alphas = 10
+
+
+# Case 1: slower runtime for (very) small alphas
+# alpha_max = 0.003471727067743962
+alpha_max = np.max(np.linalg.norm((X.T @ y).reshape(-1, 5), axis=1)) / len(y)
+alpha = alpha_max / 50
+clf = GroupLasso(fit_intercept=False,
+                 groups=5, alpha=alpha, verbose=1)
+
+t0 = time.time()
+clf.fit(X, y)
+t1 = time.time()
+
+print(f"Celer: {t1 - t0:.3f} s")
+
+t0 = time.time()
+res = group_lasso(X, y, alpha, groups=5, tol=1e-10, max_iter=10_000, check_freq=10)
+t1 = time.time()
+
+print(f"skglm gram: {t1 - t0:.3f} s")
+
+# # Case 2: slower runtime for (very) small alphas with weights
+# # alpha_max_w = 0.0001897719130007628
+# alpha_max_w = np.max(norm((X.T @ y).reshape(-1, 5) /
+#                           weights[:, None], axis=1)) / len(y)
+
+
+# alpha_ratio = 0.1
+# grid_w = np.geomspace(alpha_max_w*alpha_ratio, alpha_max_w, n_alphas)[::-1]
+# clf = GroupLasso(fit_intercept=False,
+#                  weights=weights, groups=grps, warm_start=True)
+
+# # for alpha in grid_w:
+# #     clf.alpha = alpha
+# #     t0 = time.time()
+# #     clf.fit(X, y)
+#     t1 = time.time()
+#     print(f"Finished tuning with {alpha:.2e}. Took {t1-t0:.2f} seconds!")

skglm/gram_solver.py

@@ -120,51 +120,52 @@ def group_lasso(X, y, alpha, groups, max_iter, tol, check_freq=50):
     return w
 
 
-n_samples, n_features = 1_000_000, 300
-X, y, w_star = make_correlated_data(
-    n_samples=n_samples, n_features=n_features, random_state=0)
-alpha_max = norm(X.T @ y, ord=np.inf)
-
-# Hyperparameters
-max_iter = 1000
-tol = 1e-8
-reg = 0.1
-group_size = 3
-
-alpha = alpha_max * reg / n_samples
-
-# Lasso
-print("#" * 15)
-print("Lasso")
-print("#" * 15)
-start = time()
-w = lasso(X, y, alpha, max_iter, tol)
-gram_lasso_time = time() - start
-clf_sk = Lasso(alpha, tol=tol, fit_intercept=False)
-start = time()
-clf_sk.fit(X, y)
-celer_lasso_time = time() - start
-np.testing.assert_allclose(w, clf_sk.coef_, rtol=1e-5)
-
-print("\n")
-print("Celer: %.2f" % celer_lasso_time)
-print("Gram: %.2f" % gram_lasso_time)
-print("\n")
-
-# Group Lasso
-print("#" * 15)
-print("Group Lasso")
-print("#" * 15)
-start = time()
-w = group_lasso(X, y, alpha, group_size, max_iter, tol)
-gram_group_lasso_time = time() - start
-clf_celer = GroupLasso(group_size, alpha, tol=tol, fit_intercept=False)
-start = time()
-clf_celer.fit(X, y)
-celer_group_lasso_time = time() - start
-np.testing.assert_allclose(w, clf_celer.coef_, rtol=1e-1)
-
-print("\n")
-print("Celer: %.2f" % celer_group_lasso_time)
-print("Gram: %.2f" % gram_group_lasso_time)
-print("\n")
+if __name__ == "__main__":
+    n_samples, n_features = 1_000_000, 300
+    X, y, w_star = make_correlated_data(
+        n_samples=n_samples, n_features=n_features, random_state=0)
+    alpha_max = norm(X.T @ y, ord=np.inf)
+
+    # Hyperparameters
+    max_iter = 1000
+    tol = 1e-8
+    reg = 0.1
+    group_size = 3
+
+    alpha = alpha_max * reg / n_samples
+
+    # Lasso
+    print("#" * 15)
+    print("Lasso")
+    print("#" * 15)
+    start = time()
+    w = lasso(X, y, alpha, max_iter, tol)
+    gram_lasso_time = time() - start
+    clf_sk = Lasso(alpha, tol=tol, fit_intercept=False)
+    start = time()
+    clf_sk.fit(X, y)
+    celer_lasso_time = time() - start
+    np.testing.assert_allclose(w, clf_sk.coef_, rtol=1e-5)
+
+    print("\n")
+    print("Celer: %.2f" % celer_lasso_time)
+    print("Gram: %.2f" % gram_lasso_time)
+    print("\n")
+
+    # Group Lasso
+    print("#" * 15)
+    print("Group Lasso")
+    print("#" * 15)
+    start = time()
+    w = group_lasso(X, y, alpha, group_size, max_iter, tol)
+    gram_group_lasso_time = time() - start
+    clf_celer = GroupLasso(group_size, alpha, tol=tol, fit_intercept=False)
+    start = time()
+    clf_celer.fit(X, y)
+    celer_group_lasso_time = time() - start
+    np.testing.assert_allclose(w, clf_celer.coef_, rtol=1e-1)
+
+    print("\n")
+    print("Celer: %.2f" % celer_group_lasso_time)
+    print("Gram: %.2f" % gram_group_lasso_time)
+    print("\n")