Possible to improve ranking by playing MFOG computer?

[ephriam22]

I am new to MFOG after not playing for many years. I think my rank may be around 12-15 kyu. Is it possible to play the computer on MFOG to evaluate and possibly raise my ranking. The reason that I prefer to play the computer is that I am a slow player. Thank you in advance for your help.

[jeromie]

Sure, you can improve that way. It may not be the most efficient way to get better, but this is a hobby - approach it any way that works for you!

Two caveats: You may want to consider playing all or some of your games as handicap games against a stronger setting than your current level. This will make the computer a little less likely to make odd looking errors. Also, if you transition to playing people at some point, you may find that your rating is lower than the computer indicates. In part this is because there is no universal ranking system, but you are also likely to pick up certain tactics that work against the computer but fail against opponents of the same level. (This is a risk anytime all of your games are against a single opponent, but it is exagurrated against the computer since it won't adapt to your play.)

If time limits are your main reason for avoiding human players, you might also consider correspondence games. When byo-yomi periods are measured in days, you don't have to worry about being pressured to move too quickly!

[ephriam22]

Thank you for replying jeromie. I appreciate your advice about the bias introduced by playing against the computer. I was under the impression that in view of recent advances the computer could play as well as a person. I do recall that when playing a computer many years ago it's failings were quite obvious. I will try playing people, but my experience in the past was that most wanted 15 minute games while I preferred 30 minutes or more.

By the way, could you instruct me on how to set up a computer game on MFOG so that the result is ranked. Thank you.

[Pio2001]

Hi ephriam22,
Online, on kgs, it is very easy to get a game with 25 minutes per player, with the automatch function.

In real life, most players play slower than that. Usually, a game without clock lasts for more than one hour.

[schawipp]

I would strongly recommend trying also the "DarkGo" engine on OGS with automatic handicap (just click 'Play', then 'Computer' and select the DarkGo engine and make sure, that automatic handicap is enabled; note that the handycap system is "manual" i.e. at the beginning you can move several times at once and place the handicap stones where you like). On its playing profile you can find a link with further description. According to this, it's "just" a single neural network without any lookahead functionality. This seems true, as - except for the very first move - it replies each move immediately (which may feel a bit intimidating at first glance, but once you know...). That means, it only decides its games by good shape and pattern recognition, nevertheless its current rating on OGS is 5d, which is quite something.

I sometimes play handycap games against it and it is always very illustrative, how my weak positions get punished on the spot. I never had that feeling with "pre neural-net" engines before...

[jeromie]

I was under the impression that in view of recent advances the computer could play as well as a person.

Yes, at the top level computers are better than humans. And the newer neural net based players play much more naturally than the purely Monte Carlo bots. But neural nets are often not implemented at lower levels, and a kyu level player, whether human or computer, will always make some silly mistakes. The computer will just tend to make the same sort repeatedly. (Though I've been guilty of that at times, too.)

By the way, could you instruct me on how to set up a computer game on MFOG so that the result is ranked. Thank you.

I've played a bot using the MFOG engine on KGS, but I've never actually owned the program. I can't help you with that one.

[ephriam22]

Thank you all for your informative responses. I intend to follow up on your suggestions. By the way jeromie, I found the MFOG command that ranks computer games. Thanks again.

[Mike Novack]

Using MFOG 12 efficiently when your rank is ~12-15 kyu

a) I would not worry about "ranked" just yet. The lowest MCTS level would be the 3 kyu setting so you could try taking a nine stone handicap against it. It will not play always the same way.

b) And this applies to all you try, don't forget that you can ask MCTS "why". At least try that, though whether you will be able to make heads or tails of all of the reasons in favor of a move depends on your level.

c) Much of the time you want the computer set at the level such that you will be taking 3-4 handicap stones winning about 1/3 of the games. You should not be learning the bad habit of your level when you do that.

[ephriam22]

Hi Mike Novack,
Thanks for your response. I will follow your suggestions.
I'm curious about your reference - avoiding the "bad habit of your level". Is that what happens when you play against your current (low) strength?

[Mike Novack]

Essentially yes. A player is at level "n" kyu because making the mistakes typical of the "n" level that a player of say "n-3" would not be making. But when playing against another player of level "n" that is mutual. So player one perhaps "follows" when should tenuki but player two also does not recognize this and fails to take advantage (perhaps continues in the same area).

So lets say you set the program to be about three stones stronger. It will be making mistakes typical of THAT level. That's why it isn't playing at more than three stones stronger. But that's not a problem for your learning, because you aren't at that level of mistake yet. You are still learning not to make even more basic mistakes. As you learn not to make those and begin winning too high a percentage of your games, you will be adjusting the computer to play stronger, thus always learning not to make mistakes of a higher and higher level.

High handicap games teach something else, maintaining connectivity and breking through (and how to prevent that). So you can mix in some of those. But it is the low handicap (3-4 stone) games that will be helping you learn how to approach and which joseki to choose in the context of stones already on the board.

[ephriam22]

Thank you Mike. That's very helpful.

[Shoreline]

Hi ephriam22,

I am not sure you need to play computer or bots to get the proper specific timed games. I think KGS, OGS, DGS, IGS can gets you all kinds game settings easily with human. Also I highly recommend corresponding game at DGS or OGS as well for slow games.

What bots can do better for you than human is when you need to correctly assess your playing rank. You should get correct rank at beginning and when you feel your IGS rank is lagging your real playing strength. You can assess your own true rank quickly by playing with ranked bots at DGS:
forum/viewtopic.php?f=49&t=14055

This works because DGS requires only 2 or 3 ranked games to drop or rise ranks. DGS is fastest among all go servers to adjust one's rank.

Try to play with three different DGS bots simultaneously, each with proper handicap stones and all in ranked games. At end of game, your DGS rank can jump several stones with only three games. That way, you occasionally play with DGS bots to assess your go progress whenever you feel your rank is out of sync with your real strength.

[ephriam22]

Thank you shoreline. That is very helpful information. I have not played for many years so my old ranking is likely not accurate. So I will play the 9K bot on DGS a number of times to see where I am. Thanks again! Fred

[Shoreline]

Thank you shoreline. That is very helpful information. I have not played for many years so my old ranking is likely not accurate. So I will play the 9K bot on DGS a number of times to see where I am. Thanks again! Fred

You are welcome. Once you know what is your real rank at DGS, you can get rough estimate of equivalent rank at other servers or world wide
country rank by using below table:
http://senseis.xmp.net/?RankWorldwideComparison