Hi again Jay. I have re-read some the the e-mails that you sent me, and have included a portion of them so that others that browse this site might get a better understanding of what's involved in the construction of successful net building. I hope you don't mind. Also, I included it as a reminder for those times when we get off on a tangent, sort of a starting point.
<<The new nets are set up to compare the target stock to general market conditions. No related stocks are used. All relateds are indexes.>>
<<I've tried as many as 14 relateds and the nets were not any
better than with 6 or 8. Also the bigger nets take more time to train. So keep the related index files down to 6 to 10. One more thing I've noticed is that any file used for short term filter should also be used for long term. Using certain indexes for short term and others for long term seems to confuse the training process.>>
<<The nets compare the price movement of the target stock against general market conditions. No more, no less. It does not seem to matter whether the index trends the same as the stock or not. All I have done is remove inconsistent price movement data from other company's stock, that really had nothing to do with the target stock, and replaced it with general market movements. I do believe that the index files that are used for short term filter should trend somewhat like the target stock. As far as long term filter goes, it doesn't seem to matter. The comparison is against general market movements. The proof on how accurate the predictions are, is in the report. No bad trades! No stop losses!>>
I guess my questions center about your selection of the indices. Just how do you decide which are fittest for use on your nets? Also, since the documentation is so sparse, I'd be interested in your setting for short term and long term filters. Also the concept of using an index as a long term related, even though it may not have a correlation over the long term confuses me. I know that Neuro does identify patterns, but wouldn't it confuse the net if it didn't have a correlation?
I know this is a lot to digest, and will require some time to respond to, so please take your time.
Hopefully, I'll get the N60S by the end of this week and will start re-constructing my nets, experimenting with your concepts.
<<Off Topic>> Got a several more hits on NGO this week. The combination of Neuro and NGO is awesome. |
