SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Strategies & Market Trends : Neural Nets - A tool for the 90's -- Ignore unavailable to you. Want to Upgrade?


To: Carl R. who wrote (141)6/28/1998 7:01:00 PM
From: Optim  Read Replies (1) | Respond to of 871
 
Carl,

I haven't tried Brainmaker, although it is a very popular package. PC Mag has even reviewed it.

As for the speed, with most the software today it isn't really a factor. For example in NGO I have trained 5 generations of 50 networks each, in just under 30 hours. This includes having NGO optimize the structure (Backprop, GRNN, CATN, TDNN) and the inputs (out of a possible 64). I personally like the genetic features, as it allows me to feed the network a bunch of inputs, and have it tell me which works best with my desired output.

You may want to consider that the backprop algorithm may not be the best for your predictions, although with most applications it works well. Different network structures work better with different types of problems. For example a PNN (Probabilistic Neural Network) also works well with times series (stock, futures, etc) data. NGO supports a proprietary algorithm that they call CATNN that is basically backprop with the addition of lagged data. This means the net automatically goes back in time and finds the relationships that help with todays data. This saves the user from having to feed the network lags of inputs.

I would say try a demo if they have one, or phone the company and ask if they have any evaluation deals. Most good packages offer a 30 day money back offer which you might also want to ask about.

Good luck!

Optim