Black's choice in the game loses 14% to par, according to Elf.
Enjoy!
Re: Opening problems for AI: Problem 40
Posted: Sun Aug 02, 2020 5:02 pm
by Uberdude
Pretty odd that it's black's move after the press in top left, did white tenuki and play L16? I'll just go with nice and simple c17, with the white stones around I'm not sure I really want the other common followup of e18 as
1) what's the influence doing?
2) maybe white can make messy fight
3) bots like corners.
4) I like corners.
Re: Opening problems for AI: Problem 40
Posted: Mon Aug 03, 2020 1:43 am
by Gomoto
my move:
I would play C17. I also considered candidates C3 and O3.
spoiler:
Problem 40 aka "How to cut a knights move" aka "Prohibit the cut of the knights move"
Assuming White takes sente after the corner pattern resulting from she can deal with the top left but not in a clearly advantageous way.
Conclusion: the top left looks important but it's hard to find a good way to keep sente. So perhaps playing elsewhere (invading the lower left) is best.
Re: Opening problems for AI: Problem 40
Posted: Mon Aug 03, 2020 8:29 am
by Joaz Banbeck
In the upper left, black committed to taking the outside. This is a thematic continuation.
Fujisawa jumped to the 2d line with , threatening to jump to a. But it loses 14% to par, according to Elf. O Meien peeped at b for , which loses 9½% by Elf's reckoning.
attaches underneath, and then plays the Kitani descent. plays a keima connection. The hanging connection at a is also playable. After crawls Black play the kick in the top right corner, then comes back to extend to , making eye shape.
If , Black presses White down to make small life. White can tenuki with or .
If White makes small life, the advantage of the 3-3 attachment is fairly clear. If, however, White switches to the bottom right corner, the advantage may not be so obvious.
Re: Opening problems for AI: Problem 40
Posted: Fri Aug 07, 2020 1:00 am
by Uberdude
Trust O Meien to do something wacky
Re: Opening problems for AI: Problem 40
Posted: Fri Aug 07, 2020 1:01 am
by Knotwilg
Bill, doesn't separate? It seems oddly left out of the variation. Should we assume tenuki is an option here, which I wouldn't expect.
Re: Opening problems for AI: Problem 40
Posted: Fri Aug 07, 2020 1:58 am
by Bill Spight
Knotwilg wrote:Bill, doesn't separate? It seems oddly left out of the variation. Should we assume tenuki is an option here, which I wouldn't expect.
The 3-3 got 28.3k rollouts for , which is all we are working on for the variations. The commentaries do not show plays in variations that get fewer than 1500 rollouts. got 3.4k rollouts, and then got 1.7k rollouts. The top choice for , whatever it may be, got fewer than 1.5k rollouts. It may well have been K-17, but 1) we don't know, and 2) with only 3.4k rollouts, we can't be sure of . Elf's 1.5k cutoff threshold can be annoying, but with so few rollouts we can't be confident in the choice of plays.
I think such a cutoff is wise, at least for a computer program. Look what happens when Dave runs KataGo overnight. If we give KataGo 100k rollouts after in this variation, what might the mainline be?
Re: Opening problems for AI: Problem 40
Posted: Fri Aug 07, 2020 2:11 am
by Knotwilg
Bill Spight wrote: which is all we are working on
You often refer to the GoGod Elf project, on which you seem to work with multiple people. Can you remind me who, what, why ... ?
Re: Opening problems for AI: Problem 40
Posted: Fri Aug 07, 2020 6:50 am
by Bill Spight
Knotwilg wrote:
Bill Spight wrote: which is all we are working on
You often refer to the GoGod Elf project, on which you seem to work with multiple people. Can you remind me who, what, why ... ?
I don't work on that project. I found out about the 1500 rollout threshold and the 500 rollout threshold by studying the published commentaries.
Here is their ReadMe file.
ElfOpenGo wrote:
=====================================
ELF OpenGo Professional Game Analysis
=====================================
We've used ELF OpenGo to analyze almost every professional game of Go ever played (~100k games drawn from GoGoD). We then use this analysis to evaluate the quality of professional players on a per-move basis, and not just based on who won the game. You can use our visualization tool (https://dl.fbaipublicfiles.com/elfopeng ... index.html) to compare players across eras, and to view humanity's progress in the game across time.
Importantly, you can see humanity's improvement in the game in 2016, when Go AIs came onto the scene and taught humans to play at a higher level. Also notice the harm that the large historical event of WWII did to the game.
How to understand the SGF commentaries
======================================
The UI is a tool to visualize aggregation of statistics of human games. But you can also zoom-in to the analysis for an individual game, which we provide as a commentary in SGF format. The raw data files are available at:
In every move of the game, we use ELF to find the best moves and create a tree of recommendations. We also comment with two numbers: the probability that black is winning, and the number of rollouts that ELF used to analyze that position. The more rollouts, the better ELF thinks a move is.
In the SGF, the human move is always the main line. We mark ELF’s favorite move with an exclamation point. So, if the human move is annotated with an exclamation point, then ELF thinks it is the best move. When not in the main line, a) is always the preferred move, and you will not see an exclamation point.
We retain the tree of recommendations that ELF thinks are pertinent. Specifically, we keep every move in the tree that has at least 5% of the rollouts. This means we show you a tree of every move that was considered by ELF at least 5% of the time.
ELF does not have support for dynamic komi, but we still want to bring you historical analysis. So all games are analyzed as if the komi were 7.5. For games with alternate komis, you should discount the evaluations of who is winning, though the move suggestions are still high-quality.
Selfplay analysis
=================
Besides analyses of professional games, we also have analyses of the selfplays generated throughout our training process:
We've used ELF OpenGo to analyze almost every professional game of Go ever played (~100k games drawn from GoGoD).
And for the avoidance of doubt, the GoGoD database was used with my special permission. It remains copyrighted.
I obviously had discussions on various aspects with the Facebook team, but I neither had nor sought any involvement in the AI project. I think several of the FB team have now moved on, and I have had no intimations that further work will be done. In any case, self-play has made databases otiose in that regard.
Re: Opening problems for AI: Problem 40
Posted: Fri Aug 07, 2020 3:16 pm
by ez4u
John Fairbairn wrote:
We've used ELF OpenGo to analyze almost every professional game of Go ever played (~100k games drawn from GoGoD).
And for the avoidance of doubt, the GoGoD database was used with my special permission. It remains copyrighted.
I obviously had discussions on various aspects with the Facebook team, but I neither had nor sought any involvement in the AI project. I think several of the FB team have now moved on, and I have had no intimations that further work will be done. In any case, self-play has made databases otiose in that regard.
John - How would you characterise the current state of data collection of recorded games? Are there still a lot of game records in "public" hands for you to work through or are you mainly waiting on new revelations from private sources? I think the team's claim to having analyzed substantially all the pro games ever "played" is a bit of an exaggeration, but how much of the recorded history do you think you have been able to gather together?
Re: Opening problems for AI: Problem 40
Posted: Fri Aug 07, 2020 3:32 pm
by John Fairbairn
John - How would you characterise the current state of data collection of recorded games? Are there still a lot of game records in "public" hands for you to work through or are you mainly waiting on new revelations from private sources? I think the team's claim to having analyzed substantially all the pro games ever "played" is a bit of an exaggeration, but how much of the recorded history do you think you have been able to gather together?
It was their quote not mine, but having said that I now blench as I look back at a time when Mark and I calculated that there were probably about 40,000 games available. After that faux pas I have stayed well clear of trying to estimate totals.
What I do know is that I have several hundred old Chinese games to do, several hundred early 20th century from Japan, and new Edo games (e.g. Shusaku and Shuwa and Shuho and Jowa etc - big stuff, in other words) keep popping up. I have still have sunjang baduk games to do and Ryukyuan games. I'm sure these would have all been done if Mark was still around, but I prefer working on books. Not to mention that more and more modern amateurs seem to treat old games with contumely, thinking possession of Leela on their PC makes them an expert.
So the short answer to your question has to be how long is a piece of string, I'm afraid. I'm not measuring, but I keep plodding along.
she can deal with the top left but not in a clearly advantageous way.
, threatening to jump to a. But it loses 14% to par, according to Elf. O Meien peeped at b for 
, which loses 9½% by Elf's reckoning.
plays the Kitani descent. 
plays a keima connection. The hanging connection at a is also playable. After 
crawls Black play the kick in the top right corner, then comes back to extend to 
, making eye shape. 
. 
separate? It seems oddly left out of the variation. Should we assume tenuki is an option here, which I wouldn't expect.
in this variation, what might the mainline be?