Hi,
I have written a Gamma Test package. The Gamma Test (GT) is a
non-parametric non-linear modelling tool that estimates the variance of
the noise in an input/output dataset (including time series). The GT was
recently given a rigourous mathematical proof in the Royal Society. All
the papers on this work to date can be found at Antonia Jone's (Professor
of Neural and Evolutionary computing at Cardiff University) web site see
below.....
http://www.cs.cf.ac.uk/user/Antonia.J.Jones/GammaArchive/IndexPage.htm
This work has proved to be extremely useful in the non-linear modelling
process particularly neural networks, where there is now no need for a
validation set.
Before I upload version 0.01 onto CRAN I was wondering if anyone wanted to
play about with it (who is interested in non-linear modelling, signal
processing) and give me some feed back (the more constructive criticism
the better :-)) so I can make any neccessary adjustments on the code or
returned parameters. I will be adding more tools to the package in the
near future (e.g. from a simple Gamma test, we can determine the minimal
number of data points required in the training process - this is called
the M-test, also there is a model identification procedure, which I will
be working on soon to make a fully automated procedure). Additional tools
will be added in the medium-long term future as more theory is pumped out.
Apologies in advance if I have sent this to the wrong posting place -
please do not give me a b******ing. In the past people on this mail list
have ranted at me for asking questions that I could have got from google.
Cheers,
Sam.
