SI
SI
discoversearch

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
 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext  
To: Len Giammetta who wrote (412)11/27/1998 9:30:00 AM
From: Optim   of 805
 
I think that is dependent on the stock you are watching. I have been training at least 5 nets for each stock lately for at least 12 hours each. You can get a good idea of the performance of a net by comparing them.

I vary the influence dates from 5 to 30 for S/T and 1 to 3 for the others. I like to start with a run at 12 neurons, then follow it by a training run at 18 neurons. Then I set the 'increase neurons as needed option' and train a third net. The just to compare I train a 100 neuron net. This needs at least 48 hours of training.

I also try varying inputs with all the nets. Some of the better nets I have found have required me to train (at least partially) 30-40 nets before I am happy with the results. It really ins't that bad as I use the Batch mode quite a bit. It takes about half an hour to set everything up and then I let it run for a few days.

I have also found that the Agressive trading strategy works best with most stocks. I don't like using Max Profit as I envision it as curve fitting the signal thresholds on the backtest. I would rather not optimise it so that I don't run the risk of curvefitting the training set.

Again I think everyone needs to do their own testing as what works for me might be the opposite for you. It is *very* dependant on the stock and influences.

One last observation. I notice that if I feed a net a related that is too related (highly correlated) it's performance will often degrade. This is why I have experimented with removing related on alternate runs. I don't the there is a scientific methodology to it, so I just go the trial and error route.

Optim
Report TOU ViolationShare This Post
 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext  

HomeMailHotSubjectMarksPeopleMarksKeepersSettings
Terms Of UseContact UsCopyright/IP PolicyPrivacy PolicyAbout UsFAQAdvertise on SI
© 2025 Knight Sac Media.  Data provided by Twelve Data, Alpha Vantage, and CityFALCON News