bench_lr.py

# %%
from sklearn.datasets import make_classification
from sklearn.model_selection import StratifiedShuffleSplit
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression
from sklearn.pipeline import make_pipeline
from sklearn.model_selection import cross_validate

RANDOM_SEED = 2

X, y = make_classification(
    n_samples=400,
    n_features=45,
    n_informative=10,
    n_redundant=0,
    n_repeated=0,
    n_classes=2,
    n_clusters_per_class=1,
    random_state=RANDOM_SEED,
    shuffle=False
)

sss = StratifiedShuffleSplit(
    n_splits=10, test_size=0.2, random_state=RANDOM_SEED
)

pipe = make_pipeline(
    StandardScaler(),
    LogisticRegression(
        fit_intercept=True,
        max_iter=100000,
        C=1e9,
        tol=1e-8,
        verbose=1,
        penalty='none'
    )
)

# %%
pipe.set_params(logisticregression__solver='lbfgs')
results_lbfgs = cross_validate(
    pipe, X, y,
    cv=sss,
    return_estimator=True,
    n_jobs=-1
)

# %%
pipe.set_params(logisticregression__solver='saga')
results_saga = cross_validate(
    pipe, X, y,
    cv=sss,
    return_estimator=True,
    n_jobs=-1
)

# %%
import matplotlib.pyplot as plt

for pipe_lbfgs, pipe_saga in zip(results_lbfgs['estimator'],
                                 results_saga['estimator']):
    coef_lbfgs = pipe_lbfgs[-1].coef_.ravel()
    coef_saga = pipe_saga[-1].coef_.ravel()

    plt.figure()
    plt.plot(coef_lbfgs, 'r.--', label='lbfgs')
    plt.plot(coef_saga, 'bs-.', label='saga')
    plt.title('#iterations: {} LBFGS - {} SAGA'.format(
        pipe_lbfgs[-1].n_iter_,
        pipe_saga[-1].n_iter_
    ))
    plt.legend()

# %%