Weighted linear regression

[FasterPHP]

Hi

Do you, by any chance, know where I can find a Javascript library for weighted linear regression? If not, would you consider adding it to simple-statistics?

Thanks

Marcus

[tmcw]

Ref:

• http://stat.ethz.ch/R-manual/R-patched/library/stats/html/lm.html
• https://en.wikipedia.org/wiki/Weighted_least_squares#Weighted_least_squares

[FasterPHP]

Thanks for the incredibly fast reply, but I'm not sure I understand. Is there some way to use code written in R with javascript??

[tmcw]

Sorry, not enough context; I was just adding references, since it takes a little while to figure out what weighted linear regression is. I'm not aware of any js implementations, I'll look into doing it or would accept a patch.

[FasterPHP]

I see, thanks for clarifying :)

So far, the only implementation I've found in a language I understand is this: http://www.codeproject.com/Articles/25335/An-Algorithm-for-Weighted-Linear-Regression. I've actually already converted this to Javascript, which I can supply if you like, but there are a couple of issues at present: it's not clear to me what I'm supposed to put in the array X that's passed to the Regress method; also, in its current form, I'd have to re-calculate every time a new item of data becomes available. The ideal implementation, in my opinion, would allow for a moving window of data values over which to apply the regression analysis. This would allow me to use it something like this:

var lrObj = new WeightedLinearRegression(lookbackPeriod);

var returnVal;
for (var i = 0; i < someMassiveNumber; i++) {
returnVal = lrObj.getNext(newValue, newWeight);
}

Hope this makes sense!

[tmcw]

simple-statistics is pretty committed to vanilla input; I don't plan on supporting generators, streams, etc., because these make it less readable and more complex. A more tuned implementation could be placed in stream-statistics. Though it doesn't look like online methods for linear regression can be implemented with identical behavior to offline.