Using robust LAR with 'nlinfit' or similar

Hello guys,

I have been playing with 'cftool' to determine the best non linear multiple regression of a data set. My best results are achieved when selecting a robust LAR (Least absolute residuals) regression.

Since I have several data sets, I want to use 'nlinfit' now to do the same thing. However, it seems that I cannot choose a LAR robust regression as the option 'WgtFun' of 'nlinfit' only accepts 'bisquare' (the default), 'andrews', 'cauchy', 'fair', 'huber', 'logistic', 'talwar', or 'welsch'.

Unfortunately, all these options provide worse results than LAR's.

Does anybody knows if it is possible to use LAR in 'nlinfit' (or a similar automatic routine)?

Thank you all,

Xavier

Answer 1

[Shashank Prasanna]

Have you tried generating MATLAB code for you CFTOOL fit?

In CFTOOL, after you perform your fit, click on File -> Generate Code

A MATLAB function will be generated which you can call as many times for your data and will perform the exact same fit you chose on the GUI.

Answer 2

[Xavi]

Thanks Benji! Generating the code works just great!

[Shashank Prasanna]

Please reply to 'my answer' rather than in a separate answer box. This helps keeping track of different answers and accepting the right answers that may be useful for the next person who has the same question.