Recommend me a go program, please!

Stefany93 · Post by **Stefany93** » Fri Dec 31, 2010 1:01 pm

Hello colleagues,

Insipired by my siblings who have become local chess champions by reading a lot of books and using computer programes, I have decided to kind of copy them, wow

So, I am reading a lot of books about go but I am looking for a good program, that can analyse games. Like when I load a game in it, and press some button, it would show me the correct move. I am even willing to pay for a program, but only if it okay. I hope it will help me improve my game a lot. Please recommend me some.

Thank you!
Best Regards
Stefany

hyperpape · Post by **hyperpape** » Fri Dec 31, 2010 1:59 pm

I haven't heard of anything that can do that well (actually, I'm not sure many even try--I think most programs just are there to play against). While some go programs can maintain a rank of 3-4d on KGS, using quite powerful computers (not clusters, but high end desktop machines), there are at least three points that make them less valuable as a source of analysis.

1) Many people doubt that these ratings are truly representative of the computer's strength. See the Shodan Go Bet thread for more information.

2) The best computer programs incorporate Monte Carlo analysis based on doing a lot of partially random playouts, and seeing how often a particular move wins. This means that they do not select the best move based on go knowledge. So they won't necessarily teach you good style, but more importantly, they cannot explain why a move is good. Also, they can't properly analyze games where they are substantially behind (Monte Carlo engines produce absurd moves in losing positions).

3) If your computer is not as powerful, the program will perform more poorly.

Computers might be ok as an opponent--though some people dispute that, but they're even worse for analysis and study.

I'd recommend sticking to tsumego, reviews, maybe the odd theory book, and lessons if you can afford them.

judicata · Post by **judicata** » Fri Dec 31, 2010 2:03 pm

Programs are only as strong as the engines themselves, and I don't know of any program that is as useful as a review from a player several stones stronger than you. In my unscientific estimation, any current program for suggesting specific moves won't be so useful after 12kyu or so. All that said, I still use Smart Go's "analyze game" feature on occasion, although mostly for my own amusement while I review my own game.

Game databases (with a program that can search positions) are far more useful. Smart Go (and other programs) can tell you, for example, which joseki moves are most common in pro games. I'm happy with Smart Go with the Games of Go on Disc database, though there may be better programs out there.

nagano · Post by **nagano** » Fri Dec 31, 2010 6:32 pm

Mastergo($50USD) is a good commercial program, which is updated annually for free. It can search fuseki and joseki patterns, as well as user-defined patterns. It however cannot recommend moves in the way that you describe. MoyoGo has a statistical move frequency function, which is advertised as being able to predict professional moves successfully 40% of the time. This is probably the closest to what you're looking for. It's a very nice program overall, and its search functions as a whole are better than Mastergo, but its user-defined search is a rather blunt instrument compared to Mastergo's, which allows you to add or remove any point on the board from the search pattern. MoyoGo is also now freeware. So overall MoyoGo may be the best solution for you, but not without a caveat. Firstly it is somewhat of a pain to get it to run, unless you're still using XP. Secondly, the professional games in its database were stolen from the GoGoD database. I consider this to be highly unethical, and for this reason would not download MoyoGo myself until I had lawfully purchased the GoGoD database, and I would politely ask you to do the same if you decide on MoyoGo. Besides, GoGoD is relatively inexpensive (30 euros), and has tons of material that would be difficult to find elsewhere. I will say that you should stay away from Smartgo, it tries to impress you with a lot of features, but its pattern search ability is extremely cumbersome compared to all other programs. I have an idea for a program better than all of these, but it will likely be years before I get it running.

Stefany93 · Post by **Stefany93** » Sat Jan 01, 2011 12:07 pm

Thank you very much guys. It is too bad that such great programs aren't free :-|

But still, these people deserve it

Mike Novack · Post by **Mike Novack** » Sat Jan 01, 2011 1:46 pm

hyperpape wrote:2) The best computer programs incorporate Monte Carlo analysis based on doing a lot of partially random playouts, and seeing how often a particular move wins. This means that they do not select the best move based on go knowledge. So they won't necessarily teach you good style, but more importantly, they cannot explain why a move is good. Also, they can't properly analyze games where they are substantially behind (Monte Carlo engines produce absurd moves in losing positions).

Not quite true (anymore)

All MCTS programs use the results of a large number of random playouts to select the next move (which the program thinks is best). But not all try from among all possible next moves (including strange ones). ManyFaces uses an AI engine (artifical intelligence) which based upon "go knowledge" creates a set of plausible next moves each with one or more "go reasons" behind them. It then uses MCTS as the evaluator to decide which of these is best. So if you ask it to do this, it will tell you what are the "go reasons" behind the move (feature added as of 12.021). What it can't do is give you a go reason why this move (with its go reasons) is better than other moves that also have go reasons behind them.

In other words, it can tell you it made the move "to increase the liberties of group A" but it can't tell you why that was better than a different move which if it had chosen instead would have said "to increase the liberties of group B" (can't say "in this position, saving A is bigger than saving B"). That's because the decision had the implied reason "I tried both a thousand times (Monte Carlo playouts) and won more often when I saved A".

On the second point, yes. Never disable the computer's right to resign or you'll see it make really stupid moves once the game is hopeless.

Should note here: ManyFaces is not free software and is somewhat pricey the first time you buy it. However with that price you get:
1) The right to install on multiple machines of your own. It's not a "per machine" cost but a "household" cost.
2) Interim updates are free. When a totally new version comes out owners of previous versions aren't charged full price for the new one (historically, about 1/2)

emeraldemon · Post by **emeraldemon** » Sat Jan 01, 2011 1:49 pm

If you do want something free, you can try Fuego:

http://fuego.sourceforge.net/

It's about 4 kyu on my machine, you might find it worth a try.

oren · Post by **oren** » Sat Jan 01, 2011 1:49 pm

Mike Novack wrote:Should note here: ManyFaces is not free software and is somewhat pricey the first time you buy it. However with that price you get...

Just curious, are you an official salesman for this software? You seem to promote it a lot.

Mike Novack · Post by **Mike Novack** » Sat Jan 01, 2011 3:17 pm

No, I have no relationship with Fotland (besides having owned MFOG -- among others)

However if it is any help understanding my attitude about both "free software" and for sale software.......

1) I am retired after several decades in the "cypher mines". I see nothing wrong/immoral about being paid to create software. It's how I made my living. But I wasn't writing "games" software.

2) I am old enough to have been around for the original discussions from which the "free software" movement was born. The rallying cry of which was "software for the price of a book" as opposed to "free beer" (and being "academics", to us a "book" meant the typical textbook, not a cheap paperback romance). So you could expect me to still object to any (commercial) software released only under the oppressive licenses against which that fight was undertaken but by the same token might find me objecting to what I consider strange notions of what the "free" in "free software" is supposed to mean.

With regard to MFOG
1) Everybody has to decide for his or her self whether worth the price. But as I noted, the licensing terms are reasonable (as opposed to those oppresive/unreasonable licenses of the sort we were fighting against). At the moment helping a granddaughter with buying books for college so I know what "a book costs". I also know what I spend per year on go books.
2) I also know in terms of what (were I still working) I would expect to get per hour "doing software". I should note that I also might work "free" (donating my time as an analyst) but know what that's worth.

karaklis · Post by **karaklis** » Sat Jan 01, 2011 3:29 pm

emeraldemon wrote:It's about 4 kyu on my machine, you might find it worth a try.

Not sure whether it is really 4k. I have a win ratio of 2/3 against it and am only around 7k.

Mike Novack · Post by **Mike Novack** » Sat Jan 01, 2011 6:11 pm

In talking about these programs ............

1) What machine are you running? (is it "enough" machine for the program to work properly)
2) What version of the program?
3) What time control settings?

Kataklis, that your experience differs from that of emeraldemon could be a difference in one or the other of those. That can be especially true for something like Fuego where the difference between what version is available as a Windows binary and what version available if you are compiling the latest version yourself.

karaklis · Post by **karaklis** » Sat Jan 01, 2011 7:14 pm

@Mike:

(1): two core 2.3 GHz, Win 7 (32 bit) (though Fuego only uses one core)
(2): 0.4.1.SVN
(3): There are no specific time controls to set, so I am using standard time controls. Except for the first 2-5 moves which are played instantly, it takes 10-12 secs per move. I am playing roughly at the same pace (and I know I suck at playing these blitz time controls), sometimes it takes me longer to calculate a move, but sometimes I make moves instantly (as they appear obvious to me), but in sum the used time is about the same. Engines usually profit more from short time controls.

Just played a game, here you can see whether it is 4k (Fuego was playing white):

hyperpape · Post by **hyperpape** » Sat Jan 01, 2011 7:17 pm

Mike Novack wrote:
hyperpape wrote:2) The best computer programs incorporate Monte Carlo analysis based on doing a lot of partially random playouts, and seeing how often a particular move wins. This means that they do not select the best move based on go knowledge. So they won't necessarily teach you good style, but more importantly, they cannot explain why a move is good. Also, they can't properly analyze games where they are substantially behind (Monte Carlo engines produce absurd moves in losing positions).

Not quite true (anymore)

All MCTS programs use the results of a large number of random playouts to select the next move (which the program thinks is best). But not all try from among all possible next moves (including strange ones). ManyFaces uses an AI engine (artifical intelligence) which based upon "go knowledge" creates a set of plausible next moves each with one or more "go reasons" behind them. It then uses MCTS as the evaluator to decide which of these is best. So if you ask it to do this, it will tell you what are the "go reasons" behind the move (feature added as of 12.021). What it can't do is give you a go reason why this move (with its go reasons) is better than other moves that also have go reasons behind them.

In other words, it can tell you it made the move "to increase the liberties of group A" but it can't tell you why that was better than a different move which if it had chosen instead would have said "to increase the liberties of group B" (can't say "in this position, saving A is bigger than saving B"). That's because the decision had the implied reason "I tried both a thousand times (Monte Carlo playouts) and won more often when I saved A".

This is a distinction without a difference. It can tell you some things that sound like a reason why a move is good. But it can't tell you a real reason. If MFoG is 2 dan (or 1 kyu, or whatever), it's not because it understands the purpose of moves in the way a 2 dan does it. Its level of understanding is much weaker--that's why it uses Monte Carlo techniques.

Here's an analogy: if you give me a pro game, I can give reasons why most of the moves are played. But except for easy moves, I can't give the real reasons that the moves are played because I'm not particularly good at Go. It seems as if MFoG is in the same boat--it might find 2 dan move, but not be able to give 2 dan explanations of why they are good. I'm not sure exactly how good its go knowledge is, but it has to be weaker than its playing strength for the reasons sketched above (which it really seems like you're acknowledging in your post).

P.S. I did not say Monte Carlo programs play wholly random moves in my original post. I believe they almost all employ some heuristics or other go knowledge to prune the possible moves. I take it MFoG does this more than most.

psuedome · Post by **psuedome** » Sat Jan 01, 2011 7:37 pm

Mike Novack wrote:No, I have no relationship with Fotland (besides having owned MFOG -- among others)

[admin]
From the Forum Rules (viewtopic.php?f=5&t=1006):

"1. Personal Attacks
Remember the Golden Rule: do to others what you would like to be done to you. Trolling, flaming, bashing, or otherwise verbally attacking anyone is not allowed."

-Joaz Banbeck
[/admin]

psuedome · Post by **psuedome** » Sun Jan 02, 2011 10:30 am

psuedome wrote:
Mike Novack wrote:No, I have no relationship with Fotland (besides having owned MFOG -- among others)

[admin]
From the Forum Rules (viewtopic.php?f=5&t=1006):

"1. Personal Attacks
Remember the Golden Rule: do to others what you would like to be done to you. Trolling, flaming, bashing, or otherwise verbally attacking anyone is not allowed."

-Joaz Banbeck
[/admin]

Sorry Joaz, but that wasn't a personal attack. Pointing out that "Mike Novack" is actually David Fotland in no way impugns his identity;

[admin]
It does if he claims otherwise.
There are better ways to resolve this. Just contact PM one of the admins, or report a post.
-Joaz Banbeck
[/admin]

...it simply sets the record straight. Regardless, most L19 members are already aware of this.

Life In 19x19

Recommend me a go program, please!

Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!

Re: Recommend me a go program, please!