MarketTrak Question/Comment Message

Posted By: Tom D   Date: Fri Jun 15, 2007

Title: New Model & Number of Networks

 Doubtless there have been so many advances in programming, programs, and computing speed over the past 15 years. In those mist-laden days of DOS and slow speeds I trained and ran one neural network for gold price prediction. It was amazingly excellent for quite a while, and then failed. I did not have the ability to train and run multiple nets simultaneously and still look forward with additional training for 200-400 nets and keep it all under close observation and control.
 I was dubious for quite a while about your constant research toward bettering your networks, tending to consider it as over-optimizing of the past. But you have convinced me otherwise.
 Keep up the good work!

  I think we have made some significant improvements to the old perceptron, feed forward neural networks. The nets that I now use seem to work well for this application. Thanks for writing.

Return to message board

Bag Toss: high quality, made in America, and makes a great gift.

Copyright © 1996-2018 MarketTrak LLC.  All Rights Reserved.