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: Jay Hartzok who wrote (642)	1/20/1999 8:01:00 PM
From: Bob Anderton	Read Replies (1) \| Respond to of 805

Ok Neurostock aficionados, I've observed a disturbing phenomenon that I would like you all to explain or at least corroborate. Sometimes I will have good nets with lots of indicated profit suddenly go nuts and seem to loose all their net parameters because all the buy-hold-sell indications change (much for the worse) and the indicated profit falls to near zero. At first I thought that the "net had gotten corrupted" whatever that was supposed to mean, so I would tell it to "forget" and retrain from scratch with the same data. That worked, but I was loosing all the accumulated training. I should mention that this has been happening A LOT! Like >10% of my nets per night. Then I noticed that it only happens after I download update data (I use the Neurostock site). I started saving my models on a diskette so that I could replace the hard drive copy if it got corrupted. The next time one got corrupted I copied the old diskette model back over on top of it only to see that it had the same near zero profit value. So it wasn't that the net file was corrupted. I then started copying both the net files and the data files to diskette. The next time a net crashed I was able to restore the net by copying both the net file and the data files back onto the hard drive, although the data files were now outdated. Weird. I then showed that I could open the new updated data file (the one that gives a crazy net) copy the new days of data over onto the old data file to update it, then write the updated old data file over the freshly downloaded data file and the net would be updated and still "normal" (as profitable as ever). All the data in the old data file that works and the freshly downloaded data file that doesn't work is identical when I compare them in Excel and the data plots the same in Neurostock. It has something to do with the data file but it's not the new data because I can tack the new data onto the top of the old data file and everything works fine. I hope everyone can understand this description. It's confusing for me and I know what I'm trying to say. On an unrelated note, I've noticed Neurostock changing what it claims its predictions were on earlier days even though no training has taken place since the earlier predictions were made. For example, it may say on 1/18/99 that its signal for 1/18/99 is Buy and then when you check the net again on 1/20/99 it says that it was signaling a sell on 1/18/99. To say that I find this disturbing and suspicious is somewhat understated. Anyway, I'd really like any input anyone can provide on the net crashing/data file phenomenon. Bob