repos

The microprediction stack

A collection of tiny, dependency free Python and Javascript packages. Some novel. Some SOTA.

volatility estimation timemachines streaming anomalies winning rating systems allocation streaming portfolios skaters distributional forecasting thurstone lattice order statistics precise online covariance humpday black-box optimization ice-skaters river online ML skfolio portfolio construction scikit-learn estimator API optuna raced in benchmarks
feeds plays well with core application outside package

The one-paragraph tour

Three cores do the numerics. skaters turns any stream into a full predictive distribution, one call at a time; precise keeps covariance matrices current online; thurstone computes exact order statistics of many contestants on a lattice. The applications sit one level up: timemachines (streaming anomaly detection with calibrated p-values, on skaters), winning (rating systems, on thurstone), allocation (portfolio construction, on precise and thurstone). humpday stands slightly apart: its own black-box optimizer, in Python and JavaScript, raced against Optuna, SciPy and friends. The dashed edges are the doors out: river for online ML pipelines (ice-skaters), skfolio and scikit-learn for the portfolio and estimator ecosystems.