Hi folks, I am wondering if there is appetite for a less ambitious step forward for STATISTICS-8. This was originally set up IIUC for GSoC ~8 years ago with a goal of (summarised):
"Create a new, standalone commons component for regression statistics, focusing first on linear and logistic regression." This goal raises some meaty design decisions around linear algebra and matrix functionality. If instead the goal was to handle the simpler linear regression cases, we could kick the can down the road on some of those hard design decisions and follow the approach of some other statistics modules and roll our our package-private implementations for just what is needed without committing to a public general offering of such functionality (or a third-party library like what some of the GSoC proposed implementations did). If there is interest, I have a spike along those lines which brings across some of the existing math3 legacy regression classes Simple/OLS Regression: https://github.com/apache/commons-statistics/compare/master...paulk-asert:commons-statistics:statistics8 I believe GLS and a healthy subset of MillarUpdatingRegression could come across in the same fashion. Logistic regression probably needs the meaty design choices made first, but we don't have a legacy version of that anyway. I do have some questions in the current spike around whether to make use of "Interval" and a throw vs NaN edge case, but probably best to gauge interest before getting too detailed. If there is interest, I can make the spike a PR and we can discuss what makes sense to make this reviewable and/or what else you would like to see. Cheers, Paul. --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
