<Trying to cut down training time by eliminating or restricting the
amount of price and volume data may be a mistake.> I look at building an accurate net like building a pyramid, with the most general information at it base ( does that make sense? You get the picture). After a couple hours of training with this data I think I have a fair idea, looking at the net and the verify, whether I'm on the right track. With that data it trains quick and "clean". If my net doesn't look too good I start over, modifying settings or data or both. After I feel confident that my nets are fit and accurate with the general data, I can then become more detailed and introduce additional data to the net, if need be. I can honestly say that I haven't missed an opportunity using just filters for the relateds though.
Usually, before I play a stock, per the signal, I reconstruct a net with as much relavent data as it needs. Since we're getting premature buys I have the time to do this.
Have you noticed a corrolation between your confidence ratings and the actual activity of your stocks? Have your predictions been less accurate with lower confidence?
By the way, I chickened out and took my profits on that play yesterday (sold early again). This morning Neuro signalled a "sell"... and the stock tanked(IMNR). Shew! |
