Date: Oct 18, 2013 11:15 AM
Author: Greg Heath
Subject: Re: Improving ANN results

"chaudhry " <bilal_zafar9@yahoo.com> wrote in message <l3omcv$bcu$1@newscl01ah.mathworks.com>...
> "Greg Heath" <heath@alumni.brown.edu> wrote in message <l3kj60$mlt$1@newscl01ah.mathworks.com>...
> > "Greg Heath" <heath@alumni.brown.edu> wrote in message <l3khqr$1fe$1@newscl01ah.mathworks.com>...
> > > "chaudhry " <bilal_zafar9@yahoo.com> wrote in message <l3ebuu$evs$1@newscl01ah.mathworks.com>...
> > >

> > > > How to improve ANN results by reducing error through hidden layer size, through MSE, or by using while loop?
> > >
> > > Your data is not a good learning example. (Small size, constant x(1,:), weak relationship between input and target )
> > >
> > > 1. Practice on MATLAB data (e.g., simplefit_dataset)

> >
-----SNIP

> SIR THNX ALOT.//I WILL NOW WORK UPON THAT. previously i jst generate the script and then do m alterations such as import my dataset jst alter inputs and targets.........rest every thing remain same.....what is the problm with that code

The code is too long for beginners to deal with. It puts all choices on an equal level instead of emphasizing what is important and what default inputs can, usually, ALWAYS be accepted. It prompts them to make to consider making too many choices that they shouldn't have to worry about.

On the other hand, the short examples in the help and doc documentation are too extreme in the opposite direction.

It would be preferable to

1. Improve the help and doc examples so that they yield a better understanding of what is really important, given defaults. For examples

a. Nothing is ever said about initializing the RNG so that designs can be duplicated
b. What does a result of perf = 13.72 tell the user? .... ABSOLUTELY NOTHING. Why?
Because perf is target scale dependent and needs to be compared with the average
target variance.
c. Because of a, the level of the target variances should, by default, be equal. If the
user wishes to weight some targets more than others, it can be done via the explicit
weighting input, EW, in the train and perform functions.

2. Illustrate how to obtain classification error rates directly without having to squint at a confusion matrix.

3. Give the GUI user the choice of a short or long version of command line code.

Hope this helps,

Greg