skfolio

skfolio implements allocation models as scikit-learn estimators, so the ordinary fit/predict interface, hyper-parameter search and the model_selection walk-forward splitters apply to portfolio construction — the guard against the data leakage that inflates backtests. The README groups the available families as naive (equal-weighted, inverse-volatility, random Dirichlet), convex (mean-risk, risk budgeting, maximum diversification, distributionally robust CVaR, benchmark tracker), clustering (hierarchical risk parity, hierarchical equal risk contribution, Schur complementary allocation, nested clusters optimization) and ensemble (stacking). S&P 500 price, index and factor datasets ship inside the package, so a complete example runs with no network access. Dependencies are scikit-learn >= 1.6.0, cvxpy-base, clarabel, numpy, scipy, pandas, joblib and plotly on Python >= 3.10.

The accompanying write-up is a seven-page arXiv preprint, “skfolio: Portfolio Optimization in Python” (arXiv:2507.04176, July 2025) — arXiv lists no journal reference and no peer-reviewed venue was found, so treat it as a preprint. Releases are archived on Zenodo under a citable concept DOI.

Two caveats. The library is backed by Skfolio Labs, which sells enterprise support and SLAs; the open library is complete and BSD-3-licensed, but the commercial relationship exists. And the API is pre-1.0 and moving fast — 57 releases, the current one 0.20.2 — so pin a version if you depend on specific class names. Created in December 2023, it is younger than comparable libraries and its dependent footprint is correspondingly thinner than its download count suggests.