It might have been intentionally started to be opposite normal.
| Life In 19x19 http://lifein19x19.com/ |
|
| Revisiting Go http://lifein19x19.com/viewtopic.php?f=48&t=10448 |
Page 3 of 5 |
| Author: | oren [ Thu Apr 02, 2015 9:46 am ] |
| Post subject: | Re: Revisiting Go |
It might have been intentionally started to be opposite normal.
|
|
| Author: | xed_over [ Thu Apr 02, 2015 11:08 am ] |
| Post subject: | Re: Revisiting Go |
oren wrote: It might have been intentionally started to be opposite normal. that's possible, of course, but very, very rare, and not very likely. And since it transposes so nicely, it makes comparing with other games so much easier, regardless. |
|
| Author: | oren [ Thu Apr 02, 2015 1:17 pm ] |
| Post subject: | Re: Revisiting Go |
xed_over wrote: oren wrote: It might have been intentionally started to be opposite normal. that's possible, of course, but very, very rare, and not very likely. And since it transposes so nicely, it makes comparing with other games so much easier, regardless. Yeah. Yoda has done it before, and it's nice to know from records when it does happen, since it is so rare. |
|
| Author: | Loons [ Sun Apr 05, 2015 3:17 am ] |
| Post subject: | Re: Revisiting Go |
...Clearly... ...Deliberate... As always, life seems to leave no time for go. Happily, (in just a few of the last 24 hours) I eventually finally wrote a go bot. My brother suggested such exploits could warrant a website, so I made an extremely inane blogspot (http://programmingimaging.blogspot.co.nz/). (My go bot, like everything else I do ever is just a slapping together of MATLAB image processing functions (likely making it the most expensive to own go bot in existence)). Now, I don't know what "Monte Carlo" means (Mount Carlos?), but that has not stopped me when it probably should have. |
|
| Author: | EdLee [ Sun Apr 05, 2015 3:23 am ] |
| Post subject: | |
Loons wrote: what "Monte Carlo" means Congrats on the Go bot. Some introduction to MCTS |
|
| Author: | Loons [ Sun Apr 05, 2015 3:29 am ] |
| Post subject: | Re: Revisiting Go |
Thanks for the link! Mine works by "the most primitive pure Monte Carlo method" ;p. I had a feeling that there was more interesting analysis afoot (what if it was a different move coincidentally being exchanged later that was the ACTUAL reason these particular games were won ...) |
|
| Author: | Loons [ Mon Apr 06, 2015 2:53 am ] |
| Post subject: | Re: Revisiting Go |
There is a kifu of my bot (now christened ImagingBot) playing itself over here. What's the normal solution for putting up an 11x11 SGF? With respect to the bot's play, the low sampling of possible next moves is very visible! Sometimes the bot clearly wanted to play in one part of the board, but the only option randomly assigned to it was ugly. Of course, I am probably writing my own narrative into the bot both in its code and my assessment. |
|
| Author: | Loons [ Mon Apr 06, 2015 5:43 pm ] | ||
| Post subject: | Re: Revisiting Go | ||
ImagingBot running for twelve hours ;p, at each move choosing eight alternatives and trying five games. It seems to really like nobi-ing. Currently, its random playouts consist of choosing random moves but passing instead of filling its own eyes (let's ignore pathological cases where that would be a good move) and stopping once the board doesn't change for four moves. On 19^2, that seems to take a long time (I forgot to run MATLAB's profiler over this 50 moves so I don't speak with confidence as to why). An obvious improvement would be an earlier cutoff for playouts, if I can find a useful rule. From another angle, since my sampling of moves is sparse I am going to work on adding heuristics.
|
|||
| Author: | Loons [ Tue Apr 07, 2015 10:24 pm ] | ||
| Post subject: | Re: Revisiting Go | ||
Horrible slugfests with lucky opposing errors. That's what you get when you don't practise consistently! On the bot front, I'm thinking about how to phrase heuristics in terms of basic haengma. Also, go goes surprisingly well on a checkered board with square pieces.
|
|||
| Author: | Loons [ Thu Apr 09, 2015 7:59 pm ] |
| Post subject: | Re: Revisiting Go |
In my defense, I'm quite tired. I need to shape up my reading so I don't make ridiculous slack mistakes and suddenly die (over and over). Also I might have only JUST realised one can simply link games from the KGS archives instead of reuploading them here. I spent a few moments telling MATLAB how to speak SGF, probably listening to SGFs next and then I will get back to wondering "why can Hirabot KGS 1d do 1500 playouts in the blink of an eye". (Blog link) |
|
| Author: | Loons [ Sun Apr 19, 2015 3:51 pm ] |
| Post subject: | Re: Revisiting Go |
Sometimes, you need a win. |
|
| Author: | Loons [ Sat May 16, 2015 8:11 pm ] |
| Post subject: | Re: Revisiting Go |
I guess I need to make a new account, this is cruel and unusual for everyone involved |
|
| Author: | Loons [ Wed Jun 17, 2015 4:32 pm ] |
| Post subject: | PSA |
A lot of people in 25+5*30 1k-1d on KGS "try" the triangle (when there is no discernible aji to it). The triangle fails in a simple application of slapping tesuji. The empty triangle may be the 'tougher' way to play, but as expected runs out of liberties. (Imagine some context). |
|
| Author: | Loons [ Wed Jun 17, 2015 4:50 pm ] | ||
| Post subject: | Losing to HiraBot on KGS | ||
The 1d HiraBot bots that someone is hosting on KGS are my kryptonite. I have something like a 15-20% win rate. With Hirabot as black, this opening can reliably be played, at which point I have a Takemiya-esque feeling of power on the board.I can see my attacking style is bad, with weird messy close and contact moves. I think after  I still have a chance to keep black at one eye, but it is not clear to me whether white can both contain black and prevent a second eye.(And after failing to kill I flounder around embarrassingly).
|
|||
| Author: | Loons [ Wed Jun 17, 2015 5:43 pm ] | ||
| Post subject: | Re: Revisiting Go | ||
Attacking slightly more cleanly (I would like to think), Hirabot eventually dies (n=1). I wonder what kind of opening dictionary Hirabot uses. It almost flipped it on me still, but I assume this will generally beat the bot. I wonder if other people are using bots that "are strong" but "have a weakness" to artificially inflate their ranks, play me, and then keep losing to regular people to artificially deflate my rank. My impression is that a way to beat bots like Hirabot 19^2 and Masec 9^2 is to reach a board state that the bot mis appraises, at which point the bot necessarily loses.
|
|||
| Author: | Loons [ Thu Jul 02, 2015 2:39 am ] |
| Post subject: | Re: Revisiting Go |
Back to thinking about my bot. I am not feeling especially inspired, so the problems I am thinking about probably have obvious solutions: When playing a joseki, if there is already an extension in more-or-less the right place, treat that as the joseki extension and play other parts of the joseki. How should the bot think about fuseki extensions, generally? Is Opening Theory Made Easy relevant to bots? |
|
| Author: | Loons [ Thu Jul 02, 2015 3:57 am ] |
| Post subject: | Re: Revisiting Go |
Attachment: gameGif.gif [ 1.27 MiB | Viewed 10908 times ] Since it knows joseki, its play looks surprisingly cogent. |
|
| Author: | Loons [ Wed Jul 08, 2015 2:32 am ] |
| Post subject: | Re: Revisiting Go |
I have quite a lot of go books. A list of books that I genuinely know where they are right now (ordered by when I found them in my rummage) (just dead trees) : The Master of Haengma A WAY of PLAY for the 21st CENTURY New Moves Takao's Master Play: The Style of Lee Changho Get Strong at ... Volume 6 (Tesuji) Ishida's, only volume 2? Shuko Only Move... Only Volume 2 Invincible Positional Judgement Reducing Territorial Frameworks When I thumb through ("study from") one of these, I will eventually remember reading bits before, though I certainly haven't reviewed every page of every book (I especially don't enjoy Yuan Zhou's tone, I think). As part of a concerted effort to stop investing free time in (often re-) watching clips of American Comedy Central news shows on Youtube, I am going to have a crack at these. Also secretly I want more, newer, and classick-er ones! But to justify that I need to have properly utilized these ones in recent memory. Will report back. Ooh, Positional Judgement and Reducing Territorial Frameworks were camouflaged between. Like. Frank Sargeson and Hunter S Thompson (My "book shelf" might be "vaguely themed piles on every flat surface"). |
|
| Author: | Loons [ Wed Jul 08, 2015 4:28 am ] |
| Post subject: | Re: Revisiting Go |
I crammed the first half of Master of Haengma. It could really have used a very English copy editor, though I don't know if that would have stopped haenema from appearing either. Chapter 1 consists of board positions before specific haengma (like two space haengma case 2). It is fun to investigate the board position armed with the knowledge of what shape the next move is. This is followed by 3-8 diagrams showing the correct haengma, mistakes and alternate correct haengma. Chapter 2 is a 22 problem test. Having just rocked through chapter 1, I got 11 / 22 correct, counting "viable alternatives" as correct. The problems are a board and a hint, typically pointing vaguely where on the board one should look. The ensuing answer for each problem is done the same as chapter 1's. Chapter 3 is titled "Amateur Haengma", chapter 4 "Professional Haengma" and chapter 5 "Comparing Haengma techniques". |
|
| Author: | Loons [ Thu Jul 09, 2015 5:02 am ] |
| Post subject: | Re: Revisiting Go |
Well, finished Master of Haengma. It was during the last chapter (comparing haengma) where I really started to hear what I believe the book was trying to say to me. So Amateur Haengma and Professional Haengma chapters I read exactly like the test chapter. The difference being that the amateur chapter treated situations that arose from amateur mistakes and ambitions. While reviewing the board position, I would read around a few likely looking haengma and make my pick. Sometimes I was right and my reasoning was even right, as often I would land on "this is a good idea! ...But". I was a bit frustrated that sometimes this was the right move, and sometimes this only seems like the right move and that is obviously a good move. I eventually realised that "finding the answer" was to have personally come up with all the diagrams, not to have a perfect intuition and only investigate a single correct line, which is a feeling I think I've had. The final chapter has two boards with a comparable change (subtle joseki change, or a pincer vs a jump or such) and then does its normal analysis to find each good haengma in light of the change. Master of Haengma, a large diagonal diagram wrote:
Correct answer
Black is active Black's large diagonal haengma at 1 is an interesting move. After white 2, black plays sente move at 3 and then occupies 5. This result looks active for black. |
|
| Page 3 of 5 | All times are UTC - 8 hours [ DST ] |
| Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |
|
