Life In 19x19

pnprog wrote: 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

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 >>

Mike Novack wrote: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.

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

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

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

Fair points! The important is having fun in its own way

Eventually I'll outgrow MFoG, but not soon.

Good luck with that! MFoG is 2~4dan strengths on KGS if I remember correctly.

pnprog wrote:Good luck with that! MFoG is 2~4dan strengths on KGS if I remember correctly.

I am far from certain that the MGOF 12.24 that Chaosrider has is the version that Fotland has running on KGS and there is also the matter of hardware <more CPU power> more MCTS playouts in a givne amount of time. Haven't asked him.