BGonline.org Forums

How to train a neural net

Posted By: Timothy Chow
Date: Monday, 16 July 2012, at 5:10 p.m.

In Response To: Match Analysis (Art Grater)

There's actually a serious question here, which is whether one can more effectively train a neural net by giving it more practice on its known weaknesses, or whether one should just crank up the number of matches without focusing on particular types of positions. Certainly there's a risk of overtraining on particular positions, but my sense is that blindly playing games from the starting position isn't the most effective strategy. Maybe some of the bot programmers can chime in here.

Messages In This Thread

• Match Analysis
  Giuliano Pietoso -- Saturday, 14 July 2012, at 10:42 p.m.
  • Match Analysis
    will b -- Sunday, 15 July 2012, at 12:18 a.m.
  • Match Analysis
    Timothy Chow -- Sunday, 15 July 2012, at 12:28 a.m.
  • Match Analysis
    Justin N. -- Sunday, 15 July 2012, at 12:53 a.m.
  • Match Analysis
    Bill Patterson -- Sunday, 15 July 2012, at 3:12 a.m.
  • Match Analysis
    Phil Simborg -- Sunday, 15 July 2012, at 3:44 a.m.
    • Match Analysis
      Bill Patterson -- Sunday, 15 July 2012, at 5:01 a.m.
    • Match Analysis
      Timothy Chow -- Sunday, 15 July 2012, at 6:05 p.m.
      • Match Analysis
        Casper van der Tak -- Sunday, 15 July 2012, at 7:11 p.m.
    • Match Analysis
      Art Grater -- Monday, 16 July 2012, at 7:36 a.m.
      • Match Analysis
        Henrik Bukkjaer -- Monday, 16 July 2012, at 10:24 a.m.
      • How to train a neural net
        Timothy Chow -- Monday, 16 July 2012, at 5:10 p.m.
  • Thank you
    Giuliano Pietoso -- Sunday, 15 July 2012, at 8:19 p.m.
  • Match Analysis
    Stick -- Wednesday, 18 July 2012, at 7:14 a.m.