precise

Run `npx skills add microprediction/precise --all -g -y` in your terminal to install all skills in this suite globally.

Use any precise estimator with partial_fit, for example EwaCovariance, to update the matrix in constant time per observation instead of recomputing np.cov in a rolling loop.

Call precise.suggest(X) to get a ranked shortlist of estimator classes based on measurable features of your data like dimension, conditioning, and tail behavior.

In high dimensions the Gaussian log-likelihood is dominated by unidentifiable small eigenvalues and ranks estimators below chance, so use block or Schur likelihood judges instead.

Yes. The keyed adapter wraps any estimator to consume dict observations keyed by name and maintains a live covariance over a changing universe.