| |
BGonline.org Forums
help with training on snowie/gnu
Posted By: Daniel Murphy In Response To: help with training on snowie/gnu (toppace)
Date: Thursday, 24 September 2009, at 7:43 p.m.
I aim for a very low error rate the first 5 moves. I want to use one program to do many many games with 5 moves for each player.
It's easy, with Gnubg or Snowie, to start a game, play five moves, analyze, start another game, etc. If your only aim is a "low error rate" according to your analyst, it's easy to enough to identify your "mistakes," and stop making them. Unfortunately, neither bots will play all 1st, 2nd 3rd, 4th and 5th moves correctly.
I'm not sure what good collecting a year's worth of stats on this exercise would do you for you (except that, hopefully, to show that your error rate decreases over time). The important error rate is the one you have on game day, not the average rate of a year of practice. It should be noted that rolling out of 3rd, 4th and 5th moves is a yet to be completed project, and that even in a year's worth of practice with randomly generated moves, you may never see all of the possible sequences as much as once.
I do think that some drill in nothing but openings can improve your game. But I think you should combine that with reviewing the best data we have on opening moves, and thinking about that, rather than blindly "correcting" yourself whenever a bot disagrees with your play. You might, for example, test yourself on opening responses, and then check the data and the bots to see where you may have gone wrong. You'll probably thus build up a small personal database of positions where you will not worry that your choice is not best, according to the bot, and a smaller collection of positions in which you continue to make errors, despite review and practice. You'll probably get the most out of the whole exercise by concentrating on eliminating your largest errors and on identifying position types that repeatedly give you trouble.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.