Thread Subject: Generalized Linear Model approaches in MATLAB

I need to fit a generalized linear model (more specifically a Generalized Additive Model) in the following situation: 100-d input data and 1-d output data, 1000-10000 data points, input data are NOT Gaussian distributed and show significant correlation.

I am more interested in obtaining the smooth functions of the general additive model than I am in actual predictions.

1. Will using glmfit help? How can I specify the number of smooth functions I want to estimate? I would prefer these smooth functions to be returned as linear filters.

2. Alternatively, is an Neural Networks-based radial-basis-function approach helpful?

3. Suggestions of any other functions, file exchange submissions will be greatly appreciated.

>I need to fit a generalized linear model (more specifically a Generalized
>Additive Model)
...
> 1. Will using glmfit help? How can I specify the number of smooth
> functions I want to estimate? I would prefer these smooth functions to be
> returned as linear filters.

Arvind, glmfit will not help here. Generalized linear models are a
different type of generalization than generalized additive models.
A generalized linear model relates a parametric linear function to a
transformed parameter of the response distribution. For example, it might
model the log of the mean of a response with a Poisson distribution. These
aren't nonparametric functions like GAM would provide.

It's possible that classregtree could help. It's a different approach to
nonparametric regression. I don't have any other suggestions right now
within the Statistics Toolbox.

-- Tom