enivid:
1. The guy just posted my ATC2008 EA, replacing my copyrights with his links.
2. To teach the EA - comment the line LoadKohonenMap(); and run a backtest on period you want to teach on.
3. To run the EA - uncomment the line LoadKohonenMap(); and run backtests on other periods. It's useless to backtest it on the periods that it has been taught on (I thought that's quite obvious).
4. To run the EA on demo or real - uncomment the line LoadKohonenMap(); and copy the rl.txt file from /tester/files/ to /experts/files/.
I think I understand how it works. It relies entirely on foreknowledge of whether the next bar is going to open higher or lower - is that right? So the teaching is supposed to create arrays which can be consulted as to their similarity to values derived from the current bar and its 5 predecessors. The best fit will either be from the Buy map, the Sell map or the Hold map.
All trades are closed at the start of each bar. Decision as to trade is made by consulting the map. If the map suggests the next bar will open higher, then BUY; if lower - SELL; if not sufficiently clear, no trade.
The problem I have found is that the map performs perfectly on a period where it has already been taught; but apparently the teaching does not apply anywhere else! Run it using that map on another period of time, backtest or real time, and there is no semblance of order.
There is also the snag that teaching takes a very long time as arrays totalling 45000 * 7 = 315,000 values have to be stored. 45,000 bars have to be backtested, according to my understanding.
I'd be glad of comments, especially from the original owner/author.
Malcolm