Silicon Investor (SI) -- The First Internet Community

STOCKTALK

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 : NeuroStock -- Ignore unavailable to you. Want to Upgrade?

To: Bob Anderton who wrote (540)	12/31/1998 10:06:00 AM
From: Jay Hartzok	Read Replies (2) \| Respond to of 805

Bob, It's my understanding that Neuro accepts the data in the way that the input boxes are checked. The filters are the preprocessed data S/T filter is the price minus the 50 day MA L/T filter is the 50 day MA Price and Volume are raw data inputs. If Neuro processes this data to some other form during the training process in its search for patterns, I know nothing about it. Since price is supposedly a raw data input, I used price as the input selection for the indicators so Neuro would see the data in it's true form. The only reason that I considered using a filter input for them would have been to smooth out some of the extreme fluctuations that the indicators have. I wasn't sure so I just used the raw input. Jay

To: Bob Anderton who wrote (540)	12/31/1998 12:08:00 PM
From: Optim	Respond to of 805

Hi Bob, I emailed Andrew about a year ago with a question regarding how NS preprocesses the data. His response was in regard to the genetic version I have, apparently the GA 'pieces' together indicators to find those that have some predictive value. Now I took that to mean that NS, in addition to optimizing the neural network, optimizes the preprocessing of data. That could mean that various technical indicators are tried and those that do well are weighted higher than those that don't. I don't know if the periods are adjusted in the indicators or if they are technical or statistical indicators. It's hard to say when you deal with a 'black-box' type application. He also went on to say that the Genetic process tunes the net by weeding out less predictive synapes. This is useful as it reduces the size of the network and shortens training time. It was also recommended that I run Simulated Annealing for 3 to 4 hours after the genetic run has completed. Again, these are my interpretations of what was sent to me. I think it is difficult for Andrew to reply in any detail without giving away to much on the internals of the program. Unfortunatly it makes it difficult for the users to figure out how to best use the program. Sorry to ramble! :) Happy New Year everyone. Optim