[SciPy-User] Is there anything like polyval2?

[josef.pktd at gmail.com]

On Fri, May 6, 2011 at 11:56 AM, Joseph Smidt <josephsmidt at gmail.com> wrote:
> Hi,
>
>   Matlab has a function called polyval2 that evaluates a 2D
> polynomial from a a least-squares fit of 2D data:
> http://www.mathworks.com/matlabcentral/fx_files/13719/1/content/polyfitweighted2/publishpolyfitweighted2.html
>
>    Does numpy or scipy have anything like this?  Can it be done with
> existing routines in a straight forward way?  I only see poly1d for
> one-dimensional polynomials.

I have written something like this several times (I often limit the
cross product to a simplex)

>>> nobs = 100
>>> x1 = np.random.randn(nobs)
>>> x2 = np.random.randn(nobs)

generate polynomial like a 2d np.vander:

>>> design = (x1[:,None, None]**np.arange(3)[None,:,None] * x2[:,None, None]**np.arange(3)[None,None,:]).reshape(nobs,-1)

>>> y = design.sum(1) + np.random.randn(nobs)

>>> params = np.linalg.lstsq(design, y)[0]
>>> predicted = np.dot(design, params)

Josef

>
>   Thanks.
>
> --
> ------------------------------------------------------------------------
> Joseph Smidt <josephsmidt at gmail.com>
>
> Physics and Astronomy
> 2165 Frederick Reines Hall
> Irvine, CA 92697-4575
> Office: 949-824-9025
> _______________________________________________
> SciPy-User mailing list
> SciPy-User at scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user
>