Jay... ok the fires raging!
I agree with what you're saying regarding the inputs (related stocks).
I guess when my relateds begin to exeed the 15 limit, I'll begin to "double, or triple check" the boxes, but remember the golden rule... "more data doesn't necessarily make a better, more accurate net" Sometimes bigger is not necessarily better.
In regards to another question that we've discussed regarding the target stock... In my opinion since what we're tring to predict is not only the price of the stock, but more specifically a variance that signals the entry point of either a short term or long term turnaround (or continuation of such following a hold signal), I can justify setting the target with either a short term or long term filter, hopefully that way we should get get a prediction of that variance 1-10 days into the future. Make sense? Maybe my next experiment will be to set the target with a long term filter, just to test the theory.
Just as an aside, have you given any thought to using NEURO for shorting. We'd be able to play in any market. Remember, the trend is you friend.
Just a wondering... what if we were to construct nets on all relateds to a given net, using all the targets and relateds used in the original net? It's too bad these things take so long to train.
Looks like Dave bailed on us! |
