clfs.append((
"Lightning-SVRG",
lightning_clf.SVRGClassifier(alpha=alpha,
eta=step_size,
tol=tol,
loss="log"),
sag_iter_range,
[],
[],
[],
[],
))
clfs.append((
"Lightning-SAG",
lightning_clf.SAGClassifier(alpha=alpha,
eta=step_size,
tol=tol,
loss="log"),
sag_iter_range,
[],
[],
[],
[],
))
# We keep only 200 features, to have a dense dataset,
# and compare to lightning SAG, which seems incorrect in the sparse case.
X_csc = X.tocsc()
nnz_in_each_features = X_csc.indptr[1:] - X_csc.indptr[:-1]
X = X_csc[:, np.argsort(nnz_in_each_features)[-200:]]
X = X.toarray()
print("dataset: %.3f MB" % (X.nbytes / 1e6))
if lightning_clf is not None and not fit_intercept:
alpha = 1. / C / n_samples
# compute the same step_size than in LR-sag
max_squared_sum = get_max_squared_sum(X)
step_size = get_auto_step_size(max_squared_sum, alpha, "log",
fit_intercept)
clfs.append(("Lightning-SVRG",
lightning_clf.SVRGClassifier(alpha=alpha,
eta=step_size,
tol=tol,
loss="log"), sag_iter_range, [],
[], [], []))
clfs.append(("Lightning-SAG",
lightning_clf.SAGClassifier(alpha=alpha,
eta=step_size,
tol=tol,
loss="log"), sag_iter_range, [],
[], [], []))
# We keep only 200 features, to have a dense dataset,
# and compare to lightning SAG, which seems incorrect in the sparse case.
X_csc = X.tocsc()
nnz_in_each_features = X_csc.indptr[1:] - X_csc.indptr[:-1]
X = X_csc[:, np.argsort(nnz_in_each_features)[-200:]]
X = X.toarray()
print("dataset: %.3f MB" % (X.nbytes / 1e6))
# Split training and testing. Switch train and test subset compared to
# LYRL2004 split, to have a larger training dataset.
n = 23149
X_test = X[:n, :]
classification_binary(
linear_model.RidgeClassifier(random_state=RANDOM_SEED)),
classification_binary(linear_model.RidgeClassifierCV()),
classification_binary(
linear_model.SGDClassifier(random_state=RANDOM_SEED)),
# Lightning Linear Classifiers
classification(light_clf.AdaGradClassifier(random_state=RANDOM_SEED)),
classification(light_clf.CDClassifier(random_state=RANDOM_SEED)),
classification(
light_clf.CDClassifier(
penalty="l1/l2", multiclass=True, random_state=RANDOM_SEED)),
classification(light_clf.FistaClassifier()),
classification(light_clf.FistaClassifier(multiclass=True)),
classification(light_clf.SAGAClassifier(random_state=RANDOM_SEED)),
classification(light_clf.SAGClassifier(random_state=RANDOM_SEED)),
classification(light_clf.SDCAClassifier(random_state=RANDOM_SEED)),
classification(light_clf.SGDClassifier(random_state=RANDOM_SEED)),
classification(
light_clf.SGDClassifier(multiclass=True,
random_state=RANDOM_SEED)),
classification_binary(
light_clf.AdaGradClassifier(random_state=RANDOM_SEED)),
classification_binary(
light_clf.CDClassifier(random_state=RANDOM_SEED)),
classification_binary(light_clf.FistaClassifier()),
classification_binary(
light_clf.SAGAClassifier(random_state=RANDOM_SEED)),
classification_binary(
light_clf.SAGClassifier(random_state=RANDOM_SEED)),
classification_binary(