msffs 0.1.0

mSFFS feature selection with a functional API and sklearn-compatible transformer

mSFFS

mSFFS (Modified Sequential Floating Forward Selection) is a Python library for feature selection, offering a simple functional API and a scikit-learn–compatible transformer.

Install

pip install msffs

Minimal usage (defaults)

import numpy as np
from msffs import msffs_select

X = np.array(
    [
        [0.1, 0.2, 0.3, 0.4],
        [0.2, 0.1, 0.4, 0.3],
        [0.15, 0.25, 0.35, 0.45],
        [0.9, 0.8, 0.7, 0.6],
        [0.8, 0.9, 0.6, 0.7],
        [0.85, 0.75, 0.65, 0.55],
    ]
)
y = np.array([0, 0, 0, 1, 1, 1])

result = msffs_select(X, k=2, y=y)
print(result["selected_indices"])

Scikit-learn pipeline usage

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.svm import SVC
from msffs import MSFFSSelector

pipe = Pipeline(
    steps=[
        ("scale", StandardScaler()),
        (
            "select",
            MSFFSSelector(
                k=2,
                estimator=SVC(),
                grid_params={"param_grid": {"C": [1.0], "kernel": ["linear"]}, "cv": 2},
            ),
        ),
    ]
)

X_selected = pipe.fit_transform(X, y)

API overview

• msffs_select(X, k, y, estimator=None, grid_params=None) returns a dict with:
  • selected_indices: indices of the selected features
  • selected_X: X subset to selected features (when available)
  • best_scores: cross-validated scores by subset size
  • winner_list: selected feature indices by subset size
  • best_params_list: best estimator params by subset size
• MSFFSSelector: scikit-learn compatible transformer for pipelines.

Defaults:

• Estimator: sklearn.linear_model.LogisticRegression(max_iter=1000, random_state=0)
• Grid params:
  • C: [0.1, 1.0, 10.0]
  • cv: 3

Notes

• mSFFS is supervised, so y is required.
• For very small datasets, reduce CV folds via grid_params={"cv": 2, ...}.
• Defaults stay lightweight and deterministic; pass any sklearn-compatible estimator when you need something heavier (e.g., XGBoost).

Development

pip install -e ".[dev,test]"
ruff check .
black --check .
pytest
python -m build

Citation

If you use this library in academic work, please cite:

• J. Li, D. De Ridder, D. Adhia, M. Hall, R. Mani and J. D. Deng, "Modified Feature Selection for Improved Classification of Resting-State Raw EEG Signals in Chronic Knee Pain," in IEEE Transactions on Biomedical Engineering, vol. 72, no. 5, pp. 1688-1696, May 2025, doi: 10.1109/TBME.2024.3517659.