Almeida-Pineda in RBp

["Fiona M Richardson" <F.1.Richardson@herts.ac.uk>]

==> Thanks for your input... I had a look at my graph log and the updaters
seem to be set up properly...so what I did is saved the graph log from the
demo file and loaded it into my own project...this log then works in the
same way as it did in the demo under my project. However, I have not yet
been able to determine any differences between my log and the demo log that
would mean a difference in their outputs. I am going to have a look at it
again, hopefuly I should have solved this trivial problem soon

Fiona M Richardson




"Fiona M Richardson" <F.1.Richardson@herts.ac.uk> writes:
> demo but when it comes to testing, unlike the demo graph log which shows
the
> network settling into  states during testing, my network does not seem to
> cycle and settle properly and just arrives in the correct state.

did you set the updaters properly (cycle process should be updating
the log and the network to see it settling)?  you can check who is
updating a given display by doing removeupdaters and then you'll see
the list (press cancel to not actually remove).

It's probably doing exactly the same thing behind the scenes, it might
be that your log file just isn't recording the same data. Try and change
the updater in the demo graph log to cycleprocess, it's probably set to
record from trialprocess instead just now.

Of course I might be very wrong about this. If this sorts it out,
though, tell people from the list how to solve it.

Padraic

Fiona M Richardson wrote:
>
> Hello All
>
> I have been attempting to build a copy (from the demo file that comes with
> the manual) of the xor problem using the Almeida-Pineda algorithm in RBp.
I
> am using the correct defaults file and as far as I can tell everything is
as
> it should be. However, my network trains in the same way as the one in the
> demo but when it comes to testing, unlike the demo graph log which shows
the
> network settling into  states during testing, my network does not seem to
> cycle and settle properly and just arrives in the correct state. As the
> networks are supposed to be the same they should behave in the same way,
> which as far as testing is concerned they do not. I have made comparisons
> between my project and the demo and have so far yet to identify the factor
> that may be causing the difference... is there anything anyone can think
of
> that I am not doing properly?
>
> Regards and thanks in anticipation
>
> Fiona M Richardson