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: Optim who wrote (376)	11/9/1998 3:36:00 PM
From: Jay Hartzok	Respond to of 805

Optim, Thanks. I don't do it very often, but I'll still check out the program. Jay

To: Optim who wrote (376)	11/11/1998 2:26:00 PM
From: Optim	Read Replies (3) \| Respond to of 805

Some thoughts on Annealing... After training about 20 nets on various Canadian stocks using BackPropagation, I came up with one stock that refused to train well. I tried varying the influences, adding/deleting related, etc. No luck, after 30 hours of training all I ended up with was a horizontal bar of dots in the scatter graph. So then I decided to try all over again, but I turned on Simulated Annealing in addition to BackProp. Using the exact same settings, the net managed to train well and gave a nice group of diagonal dots, both in sample and out of sample. It seems that the Annnealing process managed to 'guide' the BackProp algorithm to a better solution. This makes me wonder if I might be missing out on better networks by only using the BackProp option on most nets. My only concern with Simulated Annealing is that it is an optimization method, 'tuning' the network to fit the data. Because I don't know how NeuroStock works, I worry that by enabling Simulated Annealing I might in effect curve-fit the training set. What I have settled on for my own use (you may want to do otherwise!) is to turn on the Simulated Annealing process during the initial training. Once it is completed, I train for a further 2-3 hours using backprop. I also keep at least 2 months of data out of the net so that I can see how it has held up on data it has never seen before. I am then willing to trade the signals if the out of sample (verification period)looks good. It would really be great if we understood a bit more about the program so that we could better understand how to tweak it to get the most out of it's predictions. Also, I thought of a cool feature that could be implemented in 3.0. The ability to plot previous signals right on the chart would be great. We can see the volatility, but I have seen red volatility bars that I thought were sell signals, but Neurostock actually stayed in the trade. A plot of actual signals would show at a glance when the system bought of sold. What do you guys think? Sorry for rambling... Optim