I tried to find on NI site some other vi than the standard mmse for
linear regressions;, e.g. like local M-estimates (Cauchy or Lorentzian
like described in Numerical recipes for C).
For example I would be happy to get the slope a and intercept b of a set
of experimental data [X] [Y] with a scheme minimizing the sum of the
absolute deviations: min over a and b of sum(abs(Yi-aXi +b))).
I know it is a lot more intricate mathematically but sometimes useful
for more robust estimates without having to shoot down bad data
interactively.
Does anyone have a cute vi to do the trick, knowing one can stick around
local minima? I ran int problem trying to adpat Marquardt algorithms.
Hessian is ill-conditioned for this kind of problem (presumably)
Thanks in a
dvance.
--
Regards
Gérard D'Ans (gerard.dans@laborelec.be)
Laborelec Rodestraat,125 B-1630 Linkebeek
tel 32 2 382 0568
fax 32 2 282 0241
ULB SMA (gdans@ulb.ac.be)
Blvd F D Roosevelt CP165 B-1050 Brussels
Tel 32 2 650 2515
http://www.ulb.ac.be/polytech/laborulb/index.htm
