Leaning to use RPROP: Suggestions/References?

You work on a certain topic or data set and don't know how to start off with it using MemBrain? Not sure if your net design matches your problem or if there is room for optimization? Is it reasonable at all to approach your problem with NNs? Is MemBrain the correct tool to accomplish your task and to match your infrastructure?

These questions are best placed here!
Post Reply
MrCreosote
Posts: 55
Joined: Wed 21. Jul 2010, 18:43

Leaning to use RPROP: Suggestions/References?

Post by MrCreosote » Mon 1. Nov 2010, 18:29

I've never used RPROP and have been trying to find URL's that would explain what I need to know. Are there any URL's you would recommend?

Also, if you have any tips/tricks/hazards using RPROP, that would be helpful.

Thanks so much,
Tom

PS. With my previous platform, I was using BP w/Momentum. Sizing step size and momentum was basically trial and error, however, this is what I eventually settled on for my analysis:

1) Set momentum to 0.9 (the max value 1.0 would result in the following weighting: 50% of the previous step and 50% of the future step),
2) Then pick a learning rate that produced the smallest error after training for about 5 minutes.
3) Although a time series, do random pattern selection so that momentum acted as a "scattering perturbation" that would help deal with local minima.

Granted, issues with local minima could be encountered that could be dealt with using a massively manual approach, however, I never had the luxury of having so much time to do that. So that is how I arrived at my method.

Post Reply