Best fit funtion reprenting raw data points

[AlexFekken]

I just read that thread and was surprised that nobody suggested doing a Lagrange interpolation of all the points to get "the best fit" :-)

In my opinion Lagrange interpolation clearly shows how meaningless the idea of "the best fit" is if you do not restrict the domain of your models (and even if you do not restrict it enough, e.g. to all polynomials). Especially when you consider that you can add any number of randomly generated additional points to an existing data set (as long as all of the x values are different) and still always produce a perfect fit.