[SciPy-User] multidimensional polynomial fit

Oscar Gerardo Lazo Arjona algebraicamente@gmail....
Sun Jun 13 18:01:56 CDT 2010


David Goldsmith <d.l.goldsmith <at> gmail.com> writes:
> So the number of data _must_ equal the number of terms in the polynomial? 
(Anything else will be over- or under-determined and require either some sort of
error minimizing fit.)  What's your use case; just curious.DG

No, in general, the number of data points will be different than the number of
terms in the polynomial. For lower-dimmensional fits (in particular for
dimension 1) the number of points is expected to be larger than the number of
terms, but for higher dimmensional polynomials, the number of terms grows very
fast, and will probably be larger than the number of points.

At this point it works very slowly for dimmensions greater than 2. Adjusting a
set of 125 4-tuples to a polynomial of degree 3 takes approximately 12 seconds
on my intel-core2duo laptop. Perhaps it would be wise to implement it in some
compiled language. I am learning fotran right now, perhaps that would make it...

Yes, I will take sage objects out of an eventual contribution to numpy.

Oscar



More information about the SciPy-User mailing list