I'm trying to set up scid vs pc with stockfish 7, and have no idea how to configure the UCI
The LiquidLegends Lounge - Page 1068
| Forum Index > The Shopkeeper′s Inn |
|
killerdog
Denmark6522 Posts
I'm trying to set up scid vs pc with stockfish 7, and have no idea how to configure the UCI | ||
|
PrinceXizor
United States17713 Posts
www.twitch.tv/halfwingseen | ||
|
WaveofShadow
Canada31495 Posts
They've all been so nice | ||
|
wei2coolman
United States60033 Posts
On March 13 2016 02:31 WaveofShadow wrote: Seeing all these Syrian refugees for eye exams They've all been so nice The eyes, or the refugees? I kid, I kid. Also watching the AGA stream last night, the commentator mentioned that the Deepmind team made a single computer alphago play matches with a 2000 computer strong alphago a fuck ton of time. guess what the match break up ended up being for the 2000 computer strong alphago? Only 70% win rate for the 2000 computer strong alphago. What does this mean about the underlying strength of Alphago? It means that it's ability to read gamestates without having to do insane calculations is very fucking strong, and that the insane # of calculations AlphaGo can use after initial gamestate analysis only accounts for a +/20% increase in winrate. | ||
|
Slayer91
Ireland23470 Posts
| ||
|
zer0das
United States8519 Posts
On March 12 2016 15:26 AsmodeusXI wrote: DARKEST DUNGEON COMPLETE! Damn, those were some interesting final runs. Anyone else finished that thing? I keep getting destroyed by the final hag. I want to clear every boss first. She'll stun half my party and then heal for like 50 hp before I get to move again. I tried doing a man at arms buff group and it wasn't even close to working. :/ Man, you can carry 2000 gold per stack with a antiquitarian. I made 18k on a single run with a bunch of 1's and an antiquitarian. :> | ||
|
AsmodeusXI
United States15536 Posts
On March 13 2016 04:12 zer0das wrote: I keep getting destroyed by the final hag. I want to clear every boss first. She'll stun half my party and then heal for like 50 hp before I get to move again. I tried doing a man at arms buff group and it wasn't even close to working. :/ Man, you can carry 2000 gold per stack with a antiquitarian. I made 18k on a single run with a bunch of 1's and an antiquitarian. :> Final hag was easily the toughest of the Champion bosses for me. I think I streamed that one and it was just disgusting. I eventually bursted through her bullshit but man did that suck. | ||
|
zer0das
United States8519 Posts
| ||
|
jcarlsoniv
United States27922 Posts
+ Show Spoiler + ] | ||
|
Gahlo
United States35174 Posts
On March 13 2016 07:37 jcarlsoniv wrote: this was freaking exhausting to carry into my apartment and assemble by myself + Show Spoiler + ]Fancy shamcy. | ||
|
TheYango
United States47024 Posts
On March 13 2016 02:48 wei2coolman wrote: The eyes, or the refugees? I kid, I kid. Also watching the AGA stream last night, the commentator mentioned that the Deepmind team made a single computer alphago play matches with a 2000 computer strong alphago a fuck ton of time. guess what the match break up ended up being for the 2000 computer strong alphago? Only 70% win rate for the 2000 computer strong alphago. What does this mean about the underlying strength of Alphago? It means that it's ability to read gamestates without having to do insane calculations is very fucking strong, and that the insane # of calculations AlphaGo can use after initial gamestate analysis only accounts for a +/20% increase in winrate. A college friend of mine has been translating some discussion/blog posts from Chinese pros, and some interesting conclusions have emerged. One of the things that a lot of pros recognized in the first game was that in isolated scenarios, the AI would take "suboptimal" plays (suboptimal in the sense that a given sequence played out by the AI had a provably better alternative that would earn a couple more points at no loss, NOT in the sense of "human intuition says this is better"--just a strictly better sequence of plays). A lot of pros thus drew the false conclusion that the AI is "bad" from this. The slightly more nuanced conclusion is that the AI still lacks logical/reasoning skills in the human sense. Where humans can use logic and mathematical reasoning to prove that a particular line of play is better, the AI is still fundamentally relying on a winrate analysis from a large empirical dataset, it's incapable of *proving* anything, just choosing the line that from it's experience has the highest likelihood of winning the game. People posited that Lee Sedol drew this same conclusion going into the second game and therefore shifted his approach. Rather than trying to play an ambitious game where he did something unconventional to try and play outside of the AI's supposed experience, why not play a tighter, more conservative game where he simply tries to eke out points by beating the AI on these "suboptimal" plays? The funny part is that something became readily apparent to people in the second half of the second game--when the AI is playing a close game, these suboptimal plays disappear. Only when it was massively ahead in the first game did these "errors" occur. The reason for this seems pretty obvious in retrospect: AlphaGo is designed to optimize winrate. In a scenario where the AI is massively ahead, a single "misplay" might not even change the win percentage. In the extreme scenario, if the AI has already reached an unloseable game-state, the optimization could essentially be throwing out random moves because no one move is better than any other when victory is already guaranteed. Conversely, in a close game these small point advantages DO affect winrate meaningfully, and so the AI will pick the move that results in the highest win %--which is going to be the provably optimal sequence when given a large enough dataset. Even though the AI still lacks these human "logical reasoning" skills, this ends up not being an exploitable weakness because the deficiency only becomes apparent when the AI is already winning by a lot. The interesting part of all of this is that the idea of a player playing sloppy when they're incredibly ahead or behind and very clean when the game is close seems like an incredibly human quality. We'd expect an AI to play equally well in all situations, but the behavior of AlphaGo shows otherwise. | ||
|
Slayer91
Ireland23470 Posts
| ||
|
killerdog
Denmark6522 Posts
On March 13 2016 09:38 TheYango wrote: A college friend of mine has been translating some discussion/blog posts from Chinese pros, and some interesting conclusions have emerged. One of the things that a lot of pros recognized in the first game was that in isolated scenarios, the AI would take "suboptimal" plays (suboptimal in the sense that a given sequence played out by the AI had a provably better alternative that would earn a couple more points at no loss, NOT in the sense of "human intuition says this is better"--just a strictly better sequence of plays). A lot of pros thus drew the false conclusion that the AI is "bad" from this. The slightly more nuanced conclusion is that the AI still lacks logical/reasoning skills in the human sense. Where humans can use logic and mathematical reasoning to prove that a particular line of play is better, the AI is still fundamentally relying on a winrate analysis from a large empirical dataset, it's incapable of *proving* anything, just choosing the line that from it's experience has the highest likelihood of winning the game. People posited that Lee Sedol drew this same conclusion going into the second game and therefore shifted his approach. Rather than trying to play an ambitious game where he did something unconventional to try and play outside of the AI's supposed experience, why not play a tighter, more conservative game where he simply tries to eke out points by beating the AI on these "suboptimal" plays? The funny part is that something became readily apparent to people in the second half of the second game--when the AI is playing a close game, these suboptimal plays disappear. Only when it was massively ahead in the first game did these "errors" occur. The reason for this seems pretty obvious in retrospect: AlphaGo is designed to optimize winrate. In a scenario where the AI is massively ahead, a single "misplay" might not even change the win percentage. In the extreme scenario, if the AI has already reached an unloseable game-state, the optimization could essentially be throwing out random moves because no one move is better than any other when victory is already guaranteed. Conversely, in a close game these small point advantages DO affect winrate meaningfully, and so the AI will pick the move that results in the highest win %--which is going to be the provably optimal sequence when given a large enough dataset. Even though the AI still lacks these human "logical reasoning" skills, this ends up not being an exploitable weakness because the deficiency only becomes apparent when the AI is already winning by a lot. The interesting part of all of this is that the idea of a player playing sloppy when they're incredibly ahead or behind and very clean when the game is close seems like an incredibly human quality. We'd expect an AI to play equally well in all situations, but the behavior of AlphaGo shows otherwise. The most terrifying thing about this is the accuracy it implies alphago has at judging positions. It's taking relatively early midgame positions, and offering up large parts of it's lead in an effort to reduce complexity, because it's fully confident that this variation will lead it to winning by 1.5 points in 80 moves time or something. It was especially telling in the third game, where the commentators all thought it was close, then alphago suddenly started making "stupid" suboptimal moves which sac'd large parts of it's lead. It made the transition from "this game is close" to "i've won, lets reduce complexity" ridiculously early into the game. It's almost like a kind of showboating, like a chess player looking at a random, relatively even midgame position and saying "mate in 48." Sedol looked so crushed when it started happening that early on in the third game :o | ||
|
PrinceXizor
United States17713 Posts
| ||
|
Eppa!
Sweden4641 Posts
| ||
|
WaveofShadow
Canada31495 Posts
On March 13 2016 10:28 Eppa! wrote: Tried playing darkest dungeon a little again they really gutted healing. Is that new patch out? | ||
|
AsmodeusXI
United States15536 Posts
Yeah, the new patch is live. But I'm not sure that's true about healing. My Vestal was still doing work and I thought the Arb's healing got buffed. | ||
|
zer0das
United States8519 Posts
Edit: Although when I got a crit it was only 12 health out of 50... dunno what they mean by percent if that's the case. OH, it's the % heal increase. Interesting... | ||
|
ketchup
14521 Posts
On March 13 2016 09:46 killerdog wrote: The most terrifying thing about this is the accuracy it implies alphago has at judging positions. It's taking relatively early midgame positions, and offering up large parts of it's lead in an effort to reduce complexity, because it's fully confident that this variation will lead it to winning by 1.5 points in 80 moves time or something. It was especially telling in the third game, where the commentators all thought it was close, then alphago suddenly started making "stupid" suboptimal moves which sac'd large parts of it's lead. It made the transition from "this game is close" to "i've won, lets reduce complexity" ridiculously early into the game. It's almost like a kind of showboating, like a chess player looking at a random, relatively even midgame position and saying "mate in 48." Sedol looked so crushed when it started happening that early on in the third game :o Did you mean some other game? I don't think any of the pro commentary thought the third game was anywhere near close. Lee Sedol playing in the AlphaGo's territory on the bottom was basically desperation and also him trying to push to see what AlphaGo's limit is. I think most reviews of the third game will just look at the first 50 moves because that's where the majority of this game happened. I wonder if Lee Sedol will look at his response to that top left corner approach. It's really weird to say that a move so early on(move 15) decided the game, but it put him in such a bad spot that I don't know if he would play that same move again even if people praise the spirit behind the move. | ||
|
Eppa!
Sweden4641 Posts
| ||
| ||
]