I got my copy of this re-booted this morning. I think it's time for me to move on from Igowin, to MFG.

The biggest problem I have is the font of the MACHINE COMMENT text is almost microscopic. The "wrapper" font sizes are all fine, but the text in the machine comment box on the lower left side is almost unusable.

Does anyone know how to increase the font size on that text?

Other tips for using MFG?

Thx!

TCS

Sorry, I have to recommend not using many Faces of Go anymore.

I really liked it a few years ago. But nowadays Leela Zero is the way to go.

A self review with Leela Zero (ELF) looks for example like this:

https://lifein19x19.com/forum/viewtopic.php?f=48&t=15861

Which raises something I have not yet seen discussed.

These (new -- ie: last year or so) neural net based programs, how weak can they play? It is all well and good to discuss their capabilities to review games at a very high level.

The initial question was being asked by somebody who would be needing to take a fairly high handicap even against the weakest level of MCTS that ManyFaces 12 can play at << says 3 kyu, but I understand might actually be closer to 2 kyu >>

In other words, does anybody have experience with these neural net programs being asked to play against high handicaps? And can they be GRACEFULLY reduced in strength to play at say 1d? Because THAT is what players several kyu or weaker will need. An opponent that can be set to be several stones stronger than they are.

Maybe the CrazyStone nets trained on human play of different strengths might be very good teaching tools.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

For MFGo I don't know, but for LeelaZero, if it is too strong than it is certainly possible to use an older weight. However, older (kyu level) weights have a few shortcomings:

• they can't read ladders
• they don't count liberties, so sometimes lose semeais that a human would easily win
• sometimes, they think that an eyeless dragon is alive
• don't know many josekis, so games can become boring.

I like them in general, although if you play against a certain level many times you may encounter a lack of variety.

(For a plus on varity just enter a different fuseki and start play from there)

Gomoto:

MFG has as an overwhelming advantage: the fact that I already have it!

But that raises an interesting question. Are all these zippy-whamo AI programs free? That would surprise me if true, but I have no idea.

Mike Novack:

Excellent point. I've goosed my skill up to a reliable 9 kyu, but I don't know that I would enjoy playing with a 9-stone handicap against a MACHINE. It just feels wrong, somehow.

It sounds like MFG may be a really bad choice for people a lot stronger than I am...

If MFG can play close to 3 kyu, that should serve my needs for quite a while. In any event, I've currently settled on using it mostly for reviewing pro games.

Thx,

TCS

Leela Zero is free indeed.

No, MFOG 12 can play 1d. The next level down (still MCTS) is 3kyu. The 6 kyu level and below is not MCTS. You could try your MFOG at level 3 kyu taking 6 stones or level 6kyu taking 3 stones.

Three stones does not distort the play as higher handicaps do.

You CAN as MFOG "why", in which mode it will list the various things a move is doing. Unlike other MCTS programs, MFOG levels 1d and 3 kyu use MCTS to pick between a set of plausible moves, that set created by an AI. In other words, use MCTS to evaluate. Below that uses an AI to evaluate. Because all the moves in a plausible set got included in the set for "go reasons" MFOG can show you those reasons.

Gomoto:

REALLY??? OK, you may suck me into this AI stuff after all. Where can I download it? Or is it strictly on-line?

Mike:

Sorry, I'm getting lost in the acronyms. MCTS = ??

I've only dimly poked at the MFG "Why?" feature, but unless I can find a way to increase the font size in the MFG "comment" window, it's a feature that will be useless to me. If there's a way to change that font size, I would very much like to know what it is! I've poked around the MFG help menu a fair amount, but it appears silent on the issue.

My 3 kyu comment about the possible MFG capability was picked up from some other folks here. If it can play 1 dan, it will last even longer for me. My objective is to reach Shodan by the first of the year. My current assessment is that this will be difficult for me, but entirely achievable with sufficient attention.

Thanks to both of you for your help and indulgence!



TCS

MCTS = Monte Carlo Tree Search.

Look, you say you have MFOG 12 (what are you running it on that you have trouble with font size?)

What I suggest if you want to learn from it and are now around 9 kyu
a) Set it to 6 kyu and take 3 stones. See what happens. The 6 kyu level is just the rule based AI by itself, plays solidly but not aggressive enough. If you are winning more than half the time, try 2 stones.
b) When you get to that point (win > 50% at two stones) try setting it to the 3 kyu level and take 4 stones. Try to work your way down to 2 stones.
c) If you get there, set it to 1 dan and take 3 stones. If you get to 2 stones, you will need a strongfer program.

The idea is that you want the computer opponent to be significantly stronger than you are but not too much stronger.

Haha, welcome to the new world of Go, things have changed so much since one year

First, you should have a look at the existing offer in term of Go programs: https://www.reddit.com/r/cbaduk/comments/7e05b8/go_software/ (include description and links)

Seeing that you are SDK level, I would suggest instead that you Go for the "Legacy Leela" (Leela is the Go program that LeelaZero is based on). Leela has the advantage to play 9x9 and 13x13 as well, and might be a bit more accurate in close yose for 6.5pt komi. She also should play much better on handicap games. Leela is especially recommended if your intend to use it on a not so new laptop. She comes with her own interface in an easy to install way.

Then, if you are using a somewhat semi decent desktop computer (one that has a GPU), then you have plenty of options, considering you are still not dan level:

• Leela Zero is currently the strongest bot. Even more so when running the ELF weights from Facebook. Limitations: 19x19 only, chinese rules, 7.5pt komi
• AQ is also pro strength, apparently used by Japanese pro players on their server for training. Limitations: Nvidia GPU, 19x19, japanese rules, 6.5pt komi
• Ray, in is RLO version, more or less the same limitation with LeelaZero. Also use ELF weights

The above bots may be hard to install and set up. Check out the list above, some programs package it directly to make this process easier.

I was also a fan of MFoG, but clearly, commercial Go program is just dead now. I can't find any reason one would purchase MFoG, Zen or CrazyStones nowadays.

_________________
I am the author of GoReviewPartner, a small software aimed at assisting reviewing a game of Go. Give it a try!

Mike:

I say that I have MFOG 12...because that's what the program tells me that I have! However, I know it's out of date. I can't find confirmation if my purchase (which was a long time ago), but the author/owner was kind enough to give me an access code for what I have. He asked that I not download the update until he can confirm my purchase, so I haven't.

I'm running it on a 2013 Toshiba laptop, which at the time was near the top of the "middle tier" of options available. It's hard to see that to be a problem, since it's newer than the program version I have! It has fine garden variety graphics, but I didn't buy it as a graphics machine. I need to buy a new computer for my genealogical work on Ancestry anyway, and enhanced graphics capability is one of the requirements, but that won't happen soon.

Is the text font size adjustable in MFoG?

I like your 6 kyu/3 stone suggestion for MFoG a lot. I'll definitely try that!

pnprog:

"Legacy Leela" sounds like a good option. I very much like the option to play 9 x 9 or 13 x 13. Can you set it to run Japanese rules? That's all I've ever played, so I want to stick with that for at least a while. Unless the differences are trivial.

Interesting observation about commercial Go programs. In retrospect, that was foreseeable, but I didn't foresee it!

Thanks to you both!



TCS

Yes, with a 2013 mid tier laptop, I think you have to go for the Legacy Leela, the CPU version. I am still confident it can outplay MFoG

Differences between Japanese rules and Chinese rules are trivial, as long as you play all the dame points at the end of the game. The final score might be 0.5pt different if the komi is 6.5pt instead of 7.5pt

_________________
I am the author of GoReviewPartner, a small software aimed at assisting reviewing a game of Go. Give it a try!

But that isn't (yet) Chaosrider's problem. He doesn't need a stronger MCTS program but a weaker one if taking just a few handicap stones. So it is matter of what MCTS program is weak enough without becoming erratic.

"Free software" ideology should be left out of this discussion because he already HAS the program.

And BTW, I can see a "free software" issue/debate arising with the neural net programs. A neural net program all by itself does't do much. It's behavior (what function it evaluates) depends on the DATA derived from "training". I am not at all sure that offering such data on a "not free" basis would be a violation. Any discussion of the issues raised with neural nets on the "free software" discussion forums? << my days of involvement with the free software debates was long before the internet >>

I do understand that the program strength is not the key factor. If it was only about strength, I would still be using MFoG as well.

Clearly, MFoG interface is good, at the time, I was using the "Territory display" feature a lot. But unfortunately, the interface is quite old by today's standards.

Chaosrider's mentioned some font issue that make using the interface quite hard for him, so I advice him to try the Leela who has a pretty decent interface as well, and some feature that MFoG does not have.

On a personal note, my biggest issue with MFoG was the absence of GTP implementation to experiment with the bot.
In fact, in 2015, I made a program quite similar to GoReviewPartner, that was using MFoG: I would open a SGF file from MFoG, then, my program would take control of the mouse pointer and, for each moves of the SGF, automatically action the interface to ask MFoG the follow up moves on each game position. Something like:

• click on "computer plays black"
• Wait for 30s
• click on "neither" colour for computer (makes the computer stop thinking and play its move)
• click on "computer plays white"
• Wait for 30s
• click on "neither" colour for computer (makes the computer stop thinking and play its move)
• click on "computer plays black"
• Wait for 30s
• click on "neither" colour for computer (makes the computer stop thinking and play its move)
• and so on for a few moves then:
• click on undo X time
• click on previous branch
• click on next move
• and repeat again, starting with white...

I managed to "analyse" a few games this way. But the program would easily fail this way: sometime a popup would appear, and it was hard to detect and so on. So very painful.

So when Leela was released as GTP in 2016, I was thrilled! I started to implement the same process with Leela, and Go Review Partner was born

_________________
I am the author of GoReviewPartner, a small software aimed at assisting reviewing a game of Go. Give it a try!

@pnprog - Interesting that GRP's roots go back this way!

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21

Reminds me of the bad old days of SendKeys in VBA!

Fotland confirmed that I bought MFoG in 2012, so mine is now updated and current.

pnprog:

The key issue for me isn't whether LL can beat MFoG. The key issue is whether MFoG can beat ME, and usefully teach me, which it looks like it will be able to do for at least a few more months.

Yes, I would like a better UI than MFoG, but going from any existing UI to a different UI...even if a newer and better one...involves transition costs, which I don't have a lot of motivation to incur at the moment.

I continue to be concerned that LL can't be "dumbed down" enough to he helpful to me at my current level (now 8 kyu, nibbling at 7 kyu). Also, I have a larger (but fuzzier) concern about the rapid pace of automation of Go that appears to be happening. I'm not that interested in Go as a mathematical problem; I'm more interested in it as a form of HUMAN interaction.

Mike:

Good points all. I got on the internet around 1979, on the ARPA net (which was prior to the DARPA net). If I can get something for free rather than paying for it (all other things being equal (which they never are)), that's my general preference. But I have no religious objections to paying for software. Through this and other conversations, I'm starting to see a place for paid Go software rapidly emerging, which will involve programs that humanize the UIs, rather than making the engines more powerful.

Eventually I'll outgrow MFoG, but not soon. When I do, I'll get something better. But I would much prefer to PAY for SW that can at least pretend to pass the Turing Test, as opposed to free SW that can't. One of the things I like best about BWD is that I can VERY easily imagine that Bruce is TALKING to me. And I like that a LOT.

pnprog:

The font size issue persists, but isn't entirely insurmountable. For now I'm mostly using MFoG to review professional games. The font size issue barely impacts that function. When I move on to more heavily commented games, and playing against MFoG and wanting instruction from it, we'll see. But I'm not there yet.

Your biggest issue with MFoG is a topic which holds no interest for me. So, since we're trying to meet different objectives, it's not surprising that we would make different choices! "Better or Worse" can only be rationally evaluated in the context of known and shared objectives (this applies far beyond just Go).

When I was CIO for Nevada, I debated the CIO of California on a national stage regarding the issue of open source. As the manager of about $35M/yr of "IT Stuff", the LAST thing I wanted was code that would let the engineers tinker with it...with different engineers tinkering in different ways, producing incompatible versions. Ick!

On the other hand, Open STANDARDS (eg, SGF) was and remains an excellent idea.



TCS