I don't know about Fuego, but most MCTS bots have special-case code for ladders. But this is one of those things that is extremely easy to get wrong when implementing a bot, or get it to work only in some cases and not others, etc.Loons wrote:@Emeraldemon
Sorry if I missed the answer already, but is Fuego capable of laddering stones? (Given its decisions come from random playouts...)
124. Chew (3k) vs Fuego (Bot)
-
lightvector
- Lives in sente
- Posts: 759
- Joined: Sat Jun 19, 2010 10:11 pm
- Rank: maybe 2d
- GD Posts: 0
- Has thanked: 114 times
- Been thanked: 916 times
-
emeraldemon
- Gosei
- Posts: 1744
- Joined: Sun May 02, 2010 1:33 pm
- GD Posts: 0
- KGS: greendemon
- Tygem: greendemon
- DGS: smaragdaemon
- OGS: emeraldemon
- Has thanked: 697 times
- Been thanked: 287 times
-
Chew Terr
- Gosei
- Posts: 2060
- Joined: Mon Apr 19, 2010 12:45 pm
- Rank: KGS 3k
- GD Posts: 264
- KGS: Chew
- Location: Texas
- Has thanked: 546 times
- Been thanked: 172 times
- Contact:
Re: 124. Chew (3k) vs Fuego (Bot)
Dunno if the cut was actually worthwhile, but I'm fairly convinced I won't die locally.
Someday I want to be strong enough to earn KGS[-].
-
emeraldemon
- Gosei
- Posts: 1744
- Joined: Sun May 02, 2010 1:33 pm
- GD Posts: 0
- KGS: greendemon
- Tygem: greendemon
- DGS: smaragdaemon
- OGS: emeraldemon
- Has thanked: 697 times
- Been thanked: 287 times
-
ez4u
- Oza
- Posts: 2417
- Joined: Wed Feb 23, 2011 10:15 pm
- Rank: Jp 6 dan
- GD Posts: 0
- KGS: ez4u
- Location: Tokyo, Japan
- Has thanked: 2352 times
- Been thanked: 1334 times
Re: 124. Chew (3k) vs Fuego (Bot)
emeraldemon wrote:...
did it consider any plays around 
, which would create the alternative ladder shown? If so, how did it expect White to react?
Last edited by ez4u on Tue Jun 28, 2011 2:24 am, edited 1 time in total.
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21
-
flOvermind
- Lives with ko
- Posts: 295
- Joined: Wed Apr 21, 2010 3:19 am
- Rank: EGF 4 kyu
- GD Posts: 627
- Location: Linz, Austria
- Has thanked: 21 times
- Been thanked: 43 times
-
Chew Terr
- Gosei
- Posts: 2060
- Joined: Mon Apr 19, 2010 12:45 pm
- Rank: KGS 3k
- GD Posts: 264
- KGS: Chew
- Location: Texas
- Has thanked: 546 times
- Been thanked: 172 times
- Contact:
Re: 124. Chew (3k) vs Fuego (Bot)
Okay, I suppose this buys Fuego's right-side group better shape. The only alternative I see is playing the ladder almost all the way out to get several forcing moves against my bottom middle group, but it doesn't feel worthwhile. So I'm assuming that black will not push more than one more time from each side before fixing the weak side's cut points and resuming the non-ladder game proper.
Someday I want to be strong enough to earn KGS[-].
-
emeraldemon
- Gosei
- Posts: 1744
- Joined: Sun May 02, 2010 1:33 pm
- GD Posts: 0
- KGS: greendemon
- Tygem: greendemon
- DGS: smaragdaemon
- OGS: emeraldemon
- Has thanked: 697 times
- Been thanked: 287 times
-
emeraldemon
- Gosei
- Posts: 1744
- Joined: Sun May 02, 2010 1:33 pm
- GD Posts: 0
- KGS: greendemon
- Tygem: greendemon
- DGS: smaragdaemon
- OGS: emeraldemon
- Has thanked: 697 times
- Been thanked: 287 times
-
emeraldemon
- Gosei
- Posts: 1744
- Joined: Sun May 02, 2010 1:33 pm
- GD Posts: 0
- KGS: greendemon
- Tygem: greendemon
- DGS: smaragdaemon
- OGS: emeraldemon
- Has thanked: 697 times
- Been thanked: 287 times
-
ez4u
- Oza
- Posts: 2417
- Joined: Wed Feb 23, 2011 10:15 pm
- Rank: Jp 6 dan
- GD Posts: 0
- KGS: ez4u
- Location: Tokyo, Japan
- Has thanked: 2352 times
- Been thanked: 1334 times
Re: 124. Chew (3k) vs Fuego (Bot)
Generally how many alternative "next move" candidates does fuego consider? It obviously spends most of its time on a select few, but how many candidates receive at least some analysis in each turn?emeraldemon wrote:ez4u wrote:
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21
-
emeraldemon
- Gosei
- Posts: 1744
- Joined: Sun May 02, 2010 1:33 pm
- GD Posts: 0
- KGS: greendemon
- Tygem: greendemon
- DGS: smaragdaemon
- OGS: emeraldemon
- Has thanked: 697 times
- Been thanked: 287 times
-
ez4u
- Oza
- Posts: 2417
- Joined: Wed Feb 23, 2011 10:15 pm
- Rank: Jp 6 dan
- GD Posts: 0
- KGS: ez4u
- Location: Tokyo, Japan
- Has thanked: 2352 times
- Been thanked: 1334 times
Re: 124. Chew (3k) vs Fuego (Bot)
@emeraldemon
Can you see the meaning of the 23 playouts that great majority of points receive? Is there some special pattern? Obviously 23 playouts is not going to go very far!
Can you see the meaning of the 23 playouts that great majority of points receive? Is there some special pattern? Obviously 23 playouts is not going to go very far!
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21
-
perceval
- Lives in gote
- Posts: 312
- Joined: Thu Aug 05, 2010 3:35 am
- Rank: 7K KGS
- GD Posts: 0
- KGS: tictac
- Has thanked: 52 times
- Been thanked: 41 times
Re: 124. Chew (3k) vs Fuego (Bot)
@ez4u:
my guess would be that its a default minimal value, which seems rather low:
around 20-25 is the threshold where a bernouilli distribution begin to be approximated by a normal distribution so roughly:
(here we consider that a given move give a probaility p that a random playout after the move will give a won game for B, and we want to approximate p by playing a limited number of playout )
The error is of order 0.5/sqrt(23) which is approx 0.1. This is probably the reason why they picked 23 as the default value .This seems like a rather big error bar.
It might be interesting to check if this min number of playout can be tune up and if it would affect the result.
Maybe after those min 23 playouts fuego excludes all moves with a p less than 0.4 ? BUT considering the number of candidate moves it is probable than at least one as been wrongly estimated by more than 10%. (of course fuego implementors know their jobs so we can hope they have taken that into account)
Here the final evaluation for the best move is within 0.01 of 0.5 so it would be useful to have this kind of error bars on the estimation of all points as well.
having a 1% error interval means around 2500 playout.... which is a lot (to gain a factor of 10 in accuracy you need 100 times more playout). But it needs to be compared to the in depth exploration of the many variation for the preferred move: you have roughly 300 plays looking "not interesting" so exploring them all at 2500 playout would be 750.000 play out which is still far less than playout for the top move (around 3000000 according to esmeraldemon ).
So maybe fuego will run at roughly the same speed with this value tuned up to 2500 ?
Not sure, though, that it would bring a better playing strength.
Does esmeralddemon has the time and willingness to give it a a shot for 1 move ?
Edited out the most embarassing typos
my guess would be that its a default minimal value, which seems rather low:
around 20-25 is the threshold where a bernouilli distribution begin to be approximated by a normal distribution so roughly:
(here we consider that a given move give a probaility p that a random playout after the move will give a won game for B, and we want to approximate p by playing a limited number of playout )
The error is of order 0.5/sqrt(23) which is approx 0.1. This is probably the reason why they picked 23 as the default value .This seems like a rather big error bar.
It might be interesting to check if this min number of playout can be tune up and if it would affect the result.
Maybe after those min 23 playouts fuego excludes all moves with a p less than 0.4 ? BUT considering the number of candidate moves it is probable than at least one as been wrongly estimated by more than 10%. (of course fuego implementors know their jobs so we can hope they have taken that into account)
Here the final evaluation for the best move is within 0.01 of 0.5 so it would be useful to have this kind of error bars on the estimation of all points as well.
having a 1% error interval means around 2500 playout.... which is a lot (to gain a factor of 10 in accuracy you need 100 times more playout). But it needs to be compared to the in depth exploration of the many variation for the preferred move: you have roughly 300 plays looking "not interesting" so exploring them all at 2500 playout would be 750.000 play out which is still far less than playout for the top move (around 3000000 according to esmeraldemon ).
So maybe fuego will run at roughly the same speed with this value tuned up to 2500 ?
Not sure, though, that it would bring a better playing strength.
Does esmeralddemon has the time and willingness to give it a a shot for 1 move ?
Edited out the most embarassing typos
Last edited by perceval on Fri Jul 01, 2011 7:25 am, edited 3 times in total.
In theory, there is no difference between theory and practice. In practice, there is.
-
topazg
- Tengen
- Posts: 4511
- Joined: Wed Apr 21, 2010 3:08 am
- Rank: Nebulous
- GD Posts: 918
- KGS: topazg
- Location: Chatteris, UK
- Has thanked: 1579 times
- Been thanked: 650 times
- Contact:
Re: 124. Chew (3k) vs Fuego (Bot)
If we're talking about sufficient data points to improve play beyond chance, then I suspect 1 move won't tell us muchperceval wrote:@ez4u:
my guess would be that its a default minimal value, which seems rather low:
around 20-25 is the threshold where a bernouilli distribution begin to be approximated by a normal distribution so roughly:
(here we consider that a given move give a probaility p that a randaom playout after the move will give a won game for B, and we want to approximate p by playing a limited number of playout
the error is of order 0.5/sqrt(23) which is approx 0.1. This seems is a rather big error bar.
it might be interesting to check if this min number of playout can be tune up and if it would affect the result.
having a 1% error interval means around 2500 playout.... which is a lot (to gain a factor of 10 accuracy you need 100 more playout). but it needs to be compared to the in depth exploration of the may vars: you have roughtly 300 plays looking "not interesting" so exploring them all at 2500 playout would be 750.000 play out which is still far less than playout for the top move.
So maybe fuego will run at roughtly the same speed with this value tuned up ?
Not sure, though, that it would bring a better playing strenght.
Does esmeralddemon as the time and willingness to give it a a shot for 1 move ?
