by Mark Waller
FUTURE WAVE SOFTWARE
1330 S. Gertruda Ave.
Redondo Beach, CA 90277
Phone: 310 540-5373
Product: Technical analysis package and interface capability for preprocessing data for BrainMaker
System requirements: IBM-PC compatible, EGA/VGA; no CGA, no Hercules support. 256K video
memory. At least 500K of free RAM memory; hard drive. Stock market data in CompuTrac or
MetaStock data format. Dos version 3.3 or higher.
Optional: BrainMaker neural network software (highly recommended). Math coprocessor.
Commercial price data source and downloading software.
Printer: Epson FX, EX, LQ or HewlettPackard LaserJet compatible.
Neural networks fascinate me. Anyone who studies what they are and how they work is immediately
struck by how they can be adapted to predicting stock patterns and market turns. I was convinced that
neural networks could be a valuable tool for me, so I finally bought BrainMaker and settled in for the
challenge of seeing my small portfolio grow as I maneuvered through the cycles of the market.
Months went by. I spent hours in front of the computer trying this and trying that. It wasn't that
BrainMaker was a bad program, oh no; it had lived up to my expectations. No, my problem was what I
had read in every article and advertisement dealing with neural networks— preprocessing the data. And
preprocessing sounds so straightforward, but it hides hundreds of man-hours of number-crunching
through numerous spreadsheets, all of which involves thousands of keystrokes and a mind-numbing
number of steps that have to be just right for the neural network to be set up properly for BrainMaker to
digest it. It can take several hours just to set up the training file and then several hours, in some cases, to
train the network.
It was only then I asked myself, "Hm, should I have tried another technique?" By then, of course, the
grass was two feet high and my house needed painting. I was through with neural networks! I knew that
the choice of input data was the key to success, but the work involved in setting up the data, good or bad,
was too much.