buying a strong computer go program... which one?

[TXRanger]

Hi,

Just discovered these forums and I am glad I did...

I have over the years purchased several go programs and the latest was manyfaces... which I can now beat consistently.
I see Zen19 is for sale in japan and CrazyStone everywhere. The only notes are that these are ranked 3-4d or more on KGS but they use an insane cluster of machines.

My question is how strong are these program when running with a powerful single machine? I don't plan to buy a cluster to be able to play at dan level.

I am willing to buy any of them if any of these is really at dan level when I can play "reasonably" fast (like a minute per move)

Thanks

[Mike Novack]

[quote="TXRanger"]

Let's take these out of order?

"My question is how strong are these program when running with a powerful single machine? I don't plan to buy a cluster to be able to play at dan level."

Define "powerful single machine". Sorry but a single workstation with a pair of high speed 6 core processors is more machine than some of these programs are using even in competition with each other. While at the same time being less than others running on clusters. You also need to take into account how much time alloted per move (on average) as that is equivalent to machine power. But that's absolute strength, not necessarily the same as strength realtive to a human playing under the same time constraints.

So let's define a "standard machine" as a modestly powerful laptop (or desktop)with say a 2 core 2 GHz processor. What would have been a powerful machine 5 years ago. Note that there has been progress so tody you might find a powerful desktop with say an i7-2660 processor (4 cores, 3 GHz) so about three times more powerful or even a 2600K (more than four times as powerful).

I don't know the fall off in strength as machine power decreases for most of these programs but Fotland says about 1 stone between his fast machine (6 core xeon) and a "standard machine" like my laptop. So let's suppose that's true for most of these programs, about one level per four times the crunch.



"I have over the years purchased several go programs and the latest was manyfaces... which I can now beat consistently.
Over the years? What version of MFOG are you running and on what? If you are running MFOG 12.022 on a "standard machine" you should not be able to beat it consistently if you are 1-2k. It should be playing at about 1 dan. Note that version is important as you might well be able to beat 12.021. If you haven't upgraded no reason not to since that would be free as long as you already own any version of MFOG 12,


"I see Zen19 is for sale in japan and CrazyStone everywhere. The only notes are that these are ranked 3-4d or more on KGS but they use an insane cluster of machines.
My question is how strong are these program when running with a powerful single machine? I don't plan to buy a cluster to be able to play at dan level.
I am willing to buy any of them if any of these is really at dan level when I can play "reasonably" fast (like a minute per move)"

I think you can estimate that these programs would be about 2-3 ranks weaker when running on a standard machine so either should be at least 1 dan. Because time is relative, machine vs human in certain ranges, I am not at all sure stronger at a 1 minunte control applies. As you go from 10 sec to 1 minute the time might help the human more than it helps the machine. If it were an hour the added time more likely to help the machine.


So here's what I suggest. If you are not yet at MFOG 12.022 do that upgrade since it costs you nothing. If that proves not to be a strong enough opponent you can try one of the others which I believe are a stone or two above MFOG. But if you are already playing against MFOG 12.022 and beating it consistently I believe aspects of your game are much stronger than 1-2 k.

I will add that I didn't simply take Dave Fotland's word on the strength of MFOG on my machine but took it to our local go club where a couple of our strong players tried it out at the appropriate handicaps. They concluded that it was at the advertised strength "on average" meaning weaker in strategy but stronger in fighting in both cases surprising them with a counter attacks they did not see coming and these were 7d and 5d (AGA) players.

[TXRanger]

Thanks for that concise response Mike.

I believe I have manyfaces latest version which is supposed to be 1d. I can beat it almost every time when playing even and 50% with 2stones. It seems to sometime have a strange sense of priority and make moves that doom it... I haven't seen that with zen19 or crazystones' games.

What I am looking for is really a program like zen19 or crazystone that would beat me consistently so that I can progress... If I follow your analysis, I should assume that any of these would be about 2d on my computer. I guess that answer my questions. Since zen19 is not in english (unfortunately since it seems a little better than crazystones), I guess my only option is crazystones.

Thanks!

[oren]

What I am looking for is really a program like zen19 or crazystone that would beat me consistently so that I can progress...

If you're goal is to progress, why not play people? Otherwise, it just seems you will learn how to beat zen19 or crazystone consistently.

[TXRanger]

What I am looking for is really a program like zen19 or crazystone that would beat me consistently so that I can progress...

If you're goal is to progress, why not play people? Otherwise, it just seems you will learn how to beat zen19 or crazystone consistently.

People refuse to play without correct handicap. I want to play even with 2 or 3d...

[Mike Novack]

He's right Oren. You mistake the realistic choices. It's not that he could never find a human 1-2 stones stronger than he is willing to play even "teaching games". But how often? So the choice is between rare oportunities for games against the human opponent and frequent games against the computer. From which would he learn more?

And you need to actually look at the (current) strong programs. Yes, not excatly how most humans would choose to play but ....... is that easier or harder to learn to defeat than a human? Agreed, you won't learn how to play against a human whose style is "tight" since these computer programs play very "loose", leave lots unfinished, aji all over the board for a counterattack. But there are important skills to be learned playing against that style too.

TXRanger -- I am still puzzled about your experience with MFOG. You might want to do a version check* and you never did say what hardware you were running it on. Note the discussion about the behavior of Crazy Stones on hardware of inadequate power.

*Your experience might be right for version 12.021. What was the date of your last version update?

[xed_over]

What I am looking for is really a program like zen19 or crazystone that would beat me consistently so that I can progress...

If you're goal is to progress, why not play people? Otherwise, it just seems you will learn how to beat zen19 or crazystone consistently.

People refuse to play without correct handicap. I want to play even with 2 or 3d...

play guests -- they are often sandbaggers anyway. great way to play even games against stronger players without the tilde penalty.

[oren]

People refuse to play without correct handicap. I want to play even with 2 or 3d...

ASR might be good. You can also just try L19 room and see if some friendly 2 or 3d want to play a free game.

Someone can correct me if I'm wrong, but I think you can also just set up dan level accounts on Tygem or wbaduk and start from 3d.

I just prefer playing people, so it's just some suggestions to think about.

[ez4u]

Does anyone know the update policy for crazy stone or zen? if you buy the commercial program, do they update the engine or not? I could not find any info on that at the vendors' respective sites. Since these engines are changing quite rapidly, it seems a pity to pay for them if there are no updates included.

[Mike Novack]

That's a critical question when buying (any) software because you can't judge "what is the price" unitl you know what that includes.

a) Are interim version changes free?
b) Are major upgrades free or available at reduced cost for those who have once bought the program?

I don't know what the policy is for Zen of Crazy Stone so we need to ask. For MFOG the interim version changes are free and the major upgrade have been half price for those with an earlier version. But note that you have to keep up with what the vendor's policy might be as past policy is not a guarantee for the future.

It costs nothing to ask. If anybody does get an answer for Zen of Crazy Stone should post it to this forum.

[pookpooi]

I think you can safely go for CrazyStone, or if you don't mind Japanese language, 天頂の囲碁３ (name of Zen's latest version).
The problem may be caused by KGS rating system which is usually weaker than you thought, which is why Many Faces of Go disappoint you. By the way, Many Faces of Go haven't been updated since 1/23/2011, so I bet version 13 will release soon with significant improvement.

[shapenaji]

What I am looking for is really a program like zen19 or crazystone that would beat me consistently so that I can progress...

If you're goal is to progress, why not play people? Otherwise, it just seems you will learn how to beat zen19 or crazystone consistently.

I think we may be past the point where we can say that playing bots will create critical weaknesses in your play. Certainly you may not have familiarity against certain strategies, but that's no different than if you have a strong player at your club whose style and lessons dictate a lot of your learning.

[Mike Novack]

.....
The problem may be caused by KGS rating system which is usually weaker than you thought, which is why Many Faces of Go disappoint you. By the way, Many Faces of Go haven't been updated since 1/23/2011, so I bet version 13 will release soon with significant improvement.

That is possibly the cause but I was assuming that TXRanger was being consistent describing his rank (in other words, presumably 1-2k KGS).

Look, I spent too many decdes in the cypher mines figuring out what was wrong when the programmers were stumped and "that isn't the version you think you were running" was the cause an amazing percentage of the time. When something unexpected happens best to recheck all of your assumptions. I'll even add one more question for TXRanger, "was this the first time you upgraded a version of MFOG?". The reason for that question is that the procedure isn't as simple as downloading the new version. Not automatic with the download. You need to uninstall the old and reinstall the new. Since the symptom is precisely what we would expect if running MFOG 12.021 lets rule that out? (about 1 1/2 stones difference in playing strength between 12.021 and 12.022). And we never did get confirmation about hardware.

[oren]

I think we may be past the point where we can say that playing bots will create critical weaknesses in your play. Certainly you may not have familiarity against certain strategies, but that's no different than if you have a strong player at your club whose style and lessons dictate a lot of your learning.

I'm not arguing that bots will make your play better or worse, I would just prefer to see people playing people. If people just play bots, I think we have an even harder time getting more people to play go.

[hyperpape]

Conditional on people already being interested in the game and able to find opponents, yes, it's obviously better that they play other people. But imagine what it would do for the game if all computers sold came with a built in go game.