https://bz.apache.org/bugzilla/show_bug.cgi?id=62836
--- Comment #5 from Matthias Becht <matthi...@protonmail.com> --- Hmm, I just remembered what the biggest problem was, which I forgot to mention before. There needs to be at least one more sample than there are x variables in order for the multiple linear regression to work out (or at least that's how I understand it). Commons Math throws an IllegalArgumentException is there are too few samples, which is why I had added the check for `x[0].length >= x.length`. Excel, however, still calculates something, I'm just not sure exactly what. It almost seems as if it just somehow picks which variables to use. Regarding all x's being the same: it seems that Excel does do a bit more than just return the average. If the constant is set to FALSE and values are given for new_x's, it pretends that only one point was given: (x value - constant, average y value). Using that, it can then do a simple linear regression between that point and (0, 0). I don't know what it does though when there are multiple x variables on top of that because it still calculates *something*. I also realized later that what I said about Excel silently adding the point (0,0) when the constant is set to FALSE and there is only one y value is not entirely true either. Excel actually doesn't just return the y value when there are multiple x variables, it still calculates something, but again, I'm not entirely sure *what*. I'm not entirely sure what the best solution is in this case because every option I tried caused other problems. Of course it is quite possible that there is a very obvious solution which I simply missed. Sometimes I wish that instead of making Excel more "user-friendly", the developers had just stuck to the basic mathematical rules instead, especially since hitting any of these corner cases while trying to predict something probably means you're doing something wrong anyways. -- You are receiving this mail because: You are the assignee for the bug. --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@poi.apache.org For additional commands, e-mail: dev-h...@poi.apache.org
