Implementation:Scikit learn Scikit learn BenchSAGA

Knowledge Sources	Scikit_learn Scikit-learn Docs
Domains	Machine Learning, Benchmarking
Last Updated	2026-02-08 15:00 GMT

Overview

Concrete tool for benchmarking SAGA solver performance on multinomial logistic regression provided by scikit-learn.

Description

This benchmark script compares scikit-learn's SAGA solver against Lightning's SAGA and Liblinear for logistic regression. It demonstrates the gain in using multinomial logistic regression over one-vs-rest strategies. The script evaluates on multiple datasets including 20 newsgroups, RCV1, digits, and iris, measuring log loss and training time across both binary and multinomial settings with L1 and L2 penalties.

Usage

Use this benchmark to evaluate the performance advantage of SAGA's native multinomial logistic regression over one-vs-rest approaches, and to compare solver performance across different dataset sizes and penalty types.

Code Reference

Source Location

• Repository: scikit-learn
• File: benchmarks/bench_saga.py

Signature

def fit_single(
    solver,
    X,
    y,
    penalty="l2",
    single_target=True,
    C=1,
    max_iter=10,
    skip_slow=False,
    dtype=np.float64,
)

Import

from sklearn.linear_model import LogisticRegression
from sklearn.metrics import log_loss

I/O Contract

Inputs

Name	Type	Required	Description
solver	str	Yes	Solver name: 'sag', 'saga', 'liblinear', or 'lightning'
X	array-like	Yes	Feature matrix
y	array-like	Yes	Target labels
penalty	str	No	Regularization type: 'l1' or 'l2' (default: 'l2')
single_target	bool	No	Whether to use binary classification (default: True)
C	float	No	Inverse regularization strength (default: 1)
max_iter	int	No	Maximum number of iterations (default: 10)
dtype	numpy dtype	No	Data type for arrays (default: np.float64)

Outputs

Name	Type	Description
train_score	float	Log loss on training data
test_score	float	Log loss on test data
duration	float	Wall clock time of fit
Plot	matplotlib figure	Bar plots comparing solver performance

Usage Examples

Basic Usage

from sklearn.linear_model import LogisticRegression
from sklearn.datasets import load_digits
from sklearn.model_selection import train_test_split

X, y = load_digits(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)

clf = LogisticRegression(solver='saga', penalty='l2', max_iter=100, random_state=42)
clf.fit(X_train, y_train)
print("Test accuracy:", clf.score(X_test, y_test))

Related Pages

• Principle:Scikit_learn_Scikit_learn_Online_Learning