Skip to main content.
January 19th, 2011

Tired of ads? Then register an account or login

Oakfoam makes its KGS debut

Oakfoam is a home-grown computer Go player. It is coded by Francois van Niekerk with help from Steve Kroon and is under active development. It started playing on KGS today where it currently offers to play 9×9 games with 5 min absolute time on a first-come-first-served basis.

From its first day it looks to be about 5k. Be sure to look for it in the “Computer Go” room with the account name “oakfoam” and challenge it to a game! Remember that it uses Chinese scoring and not Japanese scoring.

Links: KGS Archives, Source Code

Posted by Francois Van Niekerk in Uncategorized


This entry was posted on Wednesday, January 19th, 2011 at 5:32 pm and is filed under Uncategorized. You can follow any responses to this entry through the comments RSS 2.0 feed. You can skip to the end and leave a response. Pinging is currently not allowed.

10 Responses to “Oakfoam makes its KGS debut”

  1. Steve says:

    Anyone interested in helping with further development of Oakfoam is welcome to contact me or Francois!

  2. Chris says:

    It plays an impressively good game. My main observation (based on 3 games) is that it doesn’t resign soon enough. I know this is a human requirement rather than a computer requirement, but if you’re playing in a public forum (KGS) against human players, it’s probably worth bearing in mind.

  3. Francois Van Niekerk says:

    Thanx for the feedback Chris.

    The time to resign is a difficult to tune parameter, but I will bear this in mind, especially when Oakfoam starts playing 19×19 games on KGS. Right now I’d prefer to focus on improving strength and move the decision of when to resign into the hands of the human 😉

    Also, Oakfoam currently spends quite a bit of time on its first few moves, which seems to annoy some human players. Once an opening book is implemented, this should be solved.

  4. Chris says:

    The start lag is a little annoying, but since the time limit is fixed, I didn’t mind, as it means it gives me more time to read when I need it. It may be more annoying in a 19×19 game, especially if it gets significantly longer.

    Not sure why resign tuning is hard? Presumably at every stage you have a maximal probability of victory associated with the best move, and if this is less than some threshold (e.g. 20%), you just resign? Is it tricky because you also need to consider how far from the start (or end) the game is in terms of moves?

  5. Francois Van Niekerk says:

    At the moment, if the winrate drops to below 5% and it is not close to the beginning of the game, my program resigns.

    Raising the threshold would make my program resign sooner, but you don’t want to resign too early. There is always some noise associated with the winrate and there is also the chance the opponent will make a mistake. Yes, I could tune this so that the loss of strength is negligible now, but future improvements often change aspects of the program and would mean I’d have more things to tune, slowing progress. I will admit that the current threshold could probably do with some tuning and I will have a look at improving it at some point.

  6. Chris says:

    5%! Heh heh, that seems quite rude to me 🙂 Oakfoam has a very low estimation of his opponents!

    Just kidding.

    Anyway, I was wondering what the consensus was among the programs out there about suitable winrate for resignation? I was watching the last KGS tournament, and one of the programmers (I think of Erikabot, but I could be wrong) said it resigned at 20% winrate.

  7. Francois Van Niekerk says:

    I know Fuego also uses 5%.

    But to be honest, I haven’t seen many games that Oakfoam manages to win after the winrate goes near 20%…

  8. Chris says:

    Tell us more about the name, Oakfoam?

  9. Francois Van Niekerk says:

    Sure thing:

    1. Launch random name generator.
    2. Pick one of the output names.
    3. Make note to improve generator in future.

  10. Chris says:

    It’s a great name.

Leave a Reply

You must be logged in to post a comment.