|
Baelfael posted:I love Go, but I am absolutely terrible at it; I get crushed by people that also are just learning the game. It's very inducing. Do some of you older players have any good web resources for moving your game forward from "I know how the rules"? http://gtl.xmp.net/ Guo Juan's 1 Euro Go lectures seem pretty good too, although if youre below 15k then youre probably better off just reading GTL, sensei's, and some maybe basic books. Do problems too, thats quite important. http://www.audiogolessons.com/
|
# ¿ Jul 7, 2008 03:11 |
|
|
# ¿ Apr 29, 2024 21:20 |
|
Two Percent posted:As I understand it, the absurd amount of possible plays is the reason there's no Go AI, is this right? Because I'd like to have a computer opponent to train on every once in a while but it seems we'll have to wait for quantum computers for that. The other problem is that positional evaluation is extremely difficult. In order to decide whether one move/sequence is better than another, you need a way of evaluating the board positions they produce. In chess you can use some heuristics for this - like the value of pieces both sides are left with, number of central squares controlled and so on - but evaluating positions in Go tends to be a lot harder than this (especially during the opening). The things which make one position better than another are more abstract than they are in other games, and the reasoning here tends to be more intuitive and difficult to quantify. The quality of a board position in Go is very holistic and depends on global features such as the relationship between different groups and the overall balance of power, and this is probably more difficult to capture in a formula than the quality of a local position. Ashenai posted:There are Go AIs. They are far, far worse comparitively than chess AIs; the best one is 4 kyu, if I remember correctly, which is well shy of Master status. It's still infinitely better than most of us will ever be, though. edit: There was a 9-handicap game played between the European Go champion and the best computer Go program (running on a supercomputer) played a couple of months ago, the sgf is available here if anyone wants to watch it. The program lost, but appparently it was a close game. The pro said after that he thought the computer was approaching dan level. Nodrog fucked around with this message at 02:34 on Jul 8, 2008 |
# ¿ Jul 8, 2008 01:54 |
|
double post
|
# ¿ Jul 8, 2008 01:54 |
|
helopticor posted:Has anyone in here ever read Mathematical Go? Supposedly it gives an algorithm for perfect late endgames (which I hear from my Go playing friends are not as significant in Go as in Chess). I've studied the math (combinatorial game theory) behind it, and I'd love to hear from a Go player who read the book. The approach they take is quite cool - they treat Go as being a multi-armed bandit problem, which is a widely used framework in reenforcement learning. The idea is that youre confonted with a slot machine which has several arms, where each arm has a different reward function. You can pull one arm each turn, and when you do this you get a reward corresponding to the arm you pulled. Your task is to maximise your reward over time, which means that you have to estimate the reward function for each bandit-arm and identify which ones are 'best'. When applied to Go, they essentially treat each board position as being a 'bandit', with every possible move corresponding to an 'arm' (more technically each node of the search tree is a seperate bandit). They then use results from the literature on bandits to decide which moves are most profitable to explore (because computation time is limited, you only want to explore the moves which are likely to be good). In order to assess how 'good' each move is, they combine the UCB1 algorithm which has been studied in the context of bandits, with a monte-carlo approach where moves are evaluated by the program playing lots of games of Go against itself where that move is made, and seeing the percentage in which that move leads of victory. The result is they they can cut off a lot more unprofitable branches from the search-tree than they could with standard alpha-beta pruning, and achieve decent results. Nodrog fucked around with this message at 01:51 on Jul 22, 2008 |
# ¿ Jul 22, 2008 01:48 |
|
|
# ¿ Oct 6, 2008 08:33 |
|
Pillow Face posted:Does anyone know of a good turn-based go server which you can play through email? All the turn-based servers are blocked at my work.
|
# ¿ Oct 6, 2008 22:55 |
|
Urban Renewal posted:Not to interrupt this game but I'd like to have some GO BOOK CHAT. I am thinking of maybe picking up some books during christmas sales and I am wondering what I should get.
|
# ¿ Dec 15, 2008 08:47 |
|
rawstorm posted:For those of you you have played Go a lot, do you find it more difficult to play against completely new players than beginners who have played a few times? My theory is that since Go has so many different possibilities that many different play styles can be developed, so the play styles people develop are designed to counter other people's play styles, so when a new comer arrives more experienced Go players may struggle because the new comer might start out with a very different play style. However if you play against a player who is experienced and not too much weaker than you (say 4 stones or less) and they play in a really nonstandard way that you arent familiar with (tengen opening, dual 5-5 points etc), then it could be a struggle since you dont have a clear intuition about what to do. But even so, theyre likely to fall apart when it comes to close fighting and middle game stuff, even if theyre ahead at the end of the opening. Generally, anything that departs from 'normal' textbook play is likely to suit the stronger player since they will usually have better reading/fighting skills - thats why in handicap games, the weaker player should usually try to keep things as simple as possible, whereas white will sometimes try to complicate stuff. edit: At the 5-15k level, Attack and Defence will probably improve his play more than Kagayama would, but Kageyama's book is so charming and full of fun anecdotes that he'll probably enjoy reading it more, and its something you can keep coming back to. Nodrog fucked around with this message at 02:43 on Dec 18, 2010 |
# ¿ Dec 18, 2010 02:24 |
|
|
# ¿ Apr 29, 2024 21:20 |
|
BaconBits posted:Am I the only one who doesn't think SDKs should be giving advice to DDKs? I'm around 6k and I feel weird telling a 12k where he should put his next stone, like I'm some kind of authority on the matter. Seems like at most I could explain a life and death situation but not much else.
|
# ¿ Feb 2, 2011 18:02 |