# poly2D fit terms: secret decoder ring

hegedus

Mon, 06/27/2016 - 06:04 pm

When doing a poly2D fit of order n, how do I make the correlation to a specific term in W_Coef wave.

For example if a 3 order fit, I would get a wave with 10 points and I understand that k0 is the constant term. But how do the other points correspond to the terms in my fit?

CurveFit/NTHR=0/TBOX=768 poly2D 3, RealX /X={ImageX,ImageY} /D

Is there a standard way to assign the outpoint points in the w_coef wave?

Thanks

Andy

`displayhelptopic "poly2d"`

for the in- and output of the parameters as well as to generate the 'fit result'.

HJ

June 28, 2016 at 02:34 am - Permalink

Thanks for the insight on poly2d, that solved my immediate problem. However, the problem with solving the immediate problem is that you get a follow up problem.

I did the fit below and when I use poly2D with the coefficient wave I can get the predicted values and compare to the actuals. This is good.

Now for coefficients K0-K5 which of the polynomial term do they represent? Which are the linear terms, the squared terms, the cross term?

CurveFit/TBOX=768 poly2D 2, D268_Temperature /X={Air_Temperature,Solar_Radiation} /D

fit_D268_Temperature= poly2D(W_coef,x,y)

W_coef={61.639,-2.9922,-0.076167,0.064866,0.0026198,-1.1798e-05}

V_chisq= 3448.81;V_npnts= 728;V_numNaNs= 0;V_numINFs= 0;

V_startRow= 0;V_endRow= 727;

W_sigma={13.7,0.893,0.0142,0.0144,0.000432,3.88e-06}

Coefficient values ± one standard deviation

K0 =61.639 ± 13.7

K1 =-2.9922 ± 0.893

K2 =-0.076167 ± 0.0142

K3 =0.064866 ± 0.0144

K4 =0.0026198 ± 0.000432

K5 =-1.1798e-05 ± 3.88e-06

Thanks

August 3, 2016 at 06:02 pm - Permalink

Quoting the help file that was suggested above:

"Among coefficients for a given degree, those for terms having higher powers of X are first. Thus, poly2D returns, for a coefficient wave cw:

f(x,y) = cw[0] + cw[1]*x + cw[2]*y + cw[3]*x^2 + cw[4]*x*y + cw[5]*y^2 + ..."

So, your fit means f(x,y) = 61.639 - 2.9922*x - 0.076167*y + 0.064866*x^2 + 0.0026198*x*y - 1.1798e-5*y^2.

August 4, 2016 at 07:19 am - Permalink

Here is an idea: Can the data label for the W_Coef contain the term of the fit? Thus when I send the W_coef to a table with dimension labels it is straight forward to know which is which.

Andy

August 4, 2016 at 01:54 pm - Permalink

John Weeks

WaveMetrics, Inc.

support@wavemetrics.com

August 4, 2016 at 04:46 pm - Permalink