need a curve fitting package ... ERROR!!!

Mark Siddall mes at zoo.toronto.edu
Mon Feb 21 20:38:06 EST 1994


 dmamonit at acs.ucalgary.ca (Darwin Michael Anthony Monita) mistakenly writes:
>
>
>If you are interested in fitting a curve to a series of data
>points in a graph, this can be done using the "SMOOTH" fuction in
>Systat.  There are many ways you can ask Systat to produce a
>smooth line through your data points:
>
>DWLS (distance weighted least squates
>INVERSE (inverse squared distance)
>KERNAL
>LINEAR
>lOG
>lOWESS
>NEXPO
>NORMAL
>POLY
>POWER
>QUAD (Quadratic smoothing)
>SPLINE
>AND STEP 

CAUTION:  THIS IS NOT WHAT THE POSTER ASKED FOR!!! 

In the original post the request was for a way to fit a model function
to a set of data points, such that certain paramters in the model can be
returned for analysis.

The above does not do this, but, rather, fits model-free curves to
data so one can heuristically examine trends.

As was posted earlier by someone else, the appropriate method is to
use the NONLIN systems under the STATISTICS header in SYSTAT; specify
the model being tested:

e.g.,  TRIAL(1) = 1/1-A(EXP-(B*LEVEL))

where TRIAL(1) is a column header for the dependent variable and LEVEL
is the column header for the independent variable.  Under OPTIONS
you may specify the maximum number of iterations, as well as whether
you want to use SIMPLEX or NEWTON fitting.

when you hit GO! the software will generate the best fit of the model to
TRIAL(1) and return the values of A and B in the model above as well as
a correlation coefficient and a corrected correlation coefficient.

If you want to do it for multiple trials simply go back after each run
and edit the model statement from (for example) TRIAL(1) = ... to 
TRIAL(2) = ... etc.

Hope this helps.

Mark
mes at zoo.toronto.edu

P.S.  If you want to see what the curve looks like use Graphics and PLOT.
Somewhere in there you can specify the function of an equasion to plot.

P.P.S.  Systat is available for Mac-Artsies, as well as for DOS-weenies
like me.  Not available for Unix-gods, so far as I know.
P.P.P.S.  It ain't free and you need the manuals.  Good news is there's
only three or four, they're not too thick, you usually use only 1 or 2 of 
them most times; this in constrast to the epic tome that passes for the
SAS manuals.



More information about the Bioforum mailing list