Hi
Trying to figure out where my engine is weak, I run STS suite and I observed that STS 8 (Advancement f/g/h), STS 9 (Advancement a/b/c) is where my engine is weaker. So I started to think on how to improve that results but I come to a dilemma, the values look like are related to king safety. If I value pawns in front king, my engine will not push them.
Any ideas on how to improve this positional feature? It look to me very difficult to advance pawns if pawn shield is damaged. How strong engines deal with this?
thanks
Positional dilemma
Moderators: hgm, Rebel, chrisw
-
- Posts: 620
- Joined: Fri Feb 08, 2008 10:44 am
- Location: Madrid - Spain
-
- Posts: 2204
- Joined: Sat Jan 18, 2014 10:24 am
- Location: Andorra
Re: Positional dilemma
Hello. Do you have some examples of positions?
Daniel José - http://www.andscacs.com
-
- Posts: 620
- Joined: Fri Feb 08, 2008 10:44 am
- Location: Madrid - Spain
Re: Positional dilemma
For example, in this position:cdani wrote:Hello. Do you have some examples of positions?
[D] 2r2rk1/p4qb1/1p1ppp2/2n1n1pp/2PN4/1PN2PPP/P1QBP1K1/2RR4 w - - bm f4; id "STS(v8.0) AKPC.018"; c0 "f4=10, Be3=9, Kg1=5, Ndb5=6";
if I run, my engine choose Kg1 after some time, but if I disable king shield evaluation, it almost inmediatly think f4 and does not change the line
I must admit there are other positions I dont fully undestand, i.e.
[D] 2r3k1/p3np1p/1p1pp1p1/4b3/1PP5/P3B2P/2R2PP1/3R3K w - - bm g4; id "STS(v8.0) AKPC.020"; c0 "g4=10, Bd4=5, g3=5, Kg1=5";
It think c5 and does not change opinion, in this case even If I disable king shield eval.
-
- Posts: 2929
- Joined: Sat Jan 22, 2011 12:42 am
- Location: NL
Re: Positional dilemma
I think this is about pawn storms: knowing when to push the pawns defending your own king to help expose the enemy king.Kempelen wrote: Trying to figure out where my engine is weak, I run STS suite and I observed that STS 8 (Advancement f/g/h), STS 9 (Advancement a/b/c) is where my engine is weaker. So I started to think on how to improve that results but I come to a dilemma, the values look like are related to king safety. If I value pawns in front king, my engine will not push them.
Any ideas on how to improve this positional feature? It look to me very difficult to advance pawns if pawn shield is damaged. How strong engines deal with this?
-
- Posts: 396
- Joined: Sat May 05, 2012 2:48 pm
- Full name: Oliver Roese
Re: Positional dilemma
How do you develop your eval in general? If nobody knows, what your are trying to achive, how can they help you?
I think you would first (a) identify some features, somehow program them given some parameters and finally (b) estimate that parameters using machine learning.
[D] 2r2rk1/p4qb1/1p1ppp2/2n1n1pp/2PN4/1PN2PPP/P1QBP1K1/2RR4 w - - bm f4; id "STS(v8.0) AKPC.018"; c0 "f4=10, Be3=9, Kg1=5, Ndb5=6"
The best move is allegedly 1.f4, but your engine is afraid of that. Why was 1.f4 played in the first place? Maybe(!) to chase the knight away from its central square (I am not(!) into it, to see how it really ticks).
Can your engine (a) recognize such a outwardly tiny difference? If so, could you try to tune (b) relevant parameters?
[D] 2r3k1/p3np1p/1p1pp1p1/4b3/1PP5/P3B2P/2R2PP1/3R3K w - - bm g4; id "STS(v8.0) AKPC.020"; c0 "g4=10, Bd4=5, g3=5, Kg1=5";
Black, with a meagre pawn for the exchange, has some drawing chances. Black is relative passive, mainly due to lack of space, this is also important. Both kings are bad, this should help the attacker.
In general, white tries to open up some lines for intrusion and trade as much pieces as possible, especially the rook. Black tries to keep the position closed and keep the damned rook. He wants to activate his king.
If black could liqidate the queen side, then it could be an easy draw.
Black could survive the exchange of a single pawn, but (with passive pieces) likely not the exchange of two pawns.
In view of this, 1.c5 doesn't look that far fetched from the outside, but white could run into some fortress idea. What happens after 1..bxc5 2.bxc5 d5 and 3...Nc6 and a timely Bc7? White only option left, is to intrude over b7, furthermore black can develop counterplay. I think white messed it up somewhere.
Why was 1.g4 needed? We guess that white tries some containment strategy and that 1...Nf5 could break the blockade, therefore it is necessary for this strategy. 1.g4 in itself is a terrible move and should be left at that.
Does your engine award:
- that black lacks space and is relative passive
- that black must keep the position closed or compensate for that fact somehow, e.g. with overwhelming activity and monster king
- that black wants to keep pieces, keep the rook
- likes a centralized knight on f5 and d5
- likes an active king
- ...
I think you would first (a) identify some features, somehow program them given some parameters and finally (b) estimate that parameters using machine learning.
[D] 2r2rk1/p4qb1/1p1ppp2/2n1n1pp/2PN4/1PN2PPP/P1QBP1K1/2RR4 w - - bm f4; id "STS(v8.0) AKPC.018"; c0 "f4=10, Be3=9, Kg1=5, Ndb5=6"
The best move is allegedly 1.f4, but your engine is afraid of that. Why was 1.f4 played in the first place? Maybe(!) to chase the knight away from its central square (I am not(!) into it, to see how it really ticks).
Can your engine (a) recognize such a outwardly tiny difference? If so, could you try to tune (b) relevant parameters?
[D] 2r3k1/p3np1p/1p1pp1p1/4b3/1PP5/P3B2P/2R2PP1/3R3K w - - bm g4; id "STS(v8.0) AKPC.020"; c0 "g4=10, Bd4=5, g3=5, Kg1=5";
Black, with a meagre pawn for the exchange, has some drawing chances. Black is relative passive, mainly due to lack of space, this is also important. Both kings are bad, this should help the attacker.
In general, white tries to open up some lines for intrusion and trade as much pieces as possible, especially the rook. Black tries to keep the position closed and keep the damned rook. He wants to activate his king.
If black could liqidate the queen side, then it could be an easy draw.
Black could survive the exchange of a single pawn, but (with passive pieces) likely not the exchange of two pawns.
In view of this, 1.c5 doesn't look that far fetched from the outside, but white could run into some fortress idea. What happens after 1..bxc5 2.bxc5 d5 and 3...Nc6 and a timely Bc7? White only option left, is to intrude over b7, furthermore black can develop counterplay. I think white messed it up somewhere.
Why was 1.g4 needed? We guess that white tries some containment strategy and that 1...Nf5 could break the blockade, therefore it is necessary for this strategy. 1.g4 in itself is a terrible move and should be left at that.
Does your engine award:
- that black lacks space and is relative passive
- that black must keep the position closed or compensate for that fact somehow, e.g. with overwhelming activity and monster king
- that black wants to keep pieces, keep the rook
- likes a centralized knight on f5 and d5
- likes an active king
- ...
-
- Posts: 2204
- Joined: Sat Jan 18, 2014 10:24 am
- Location: Andorra
Re: Positional dilemma
The first position doing f4 is difficult also for Andscacs. For the second has no problems as the pawn shelter and king safety stuff is much diminished for the endgames.
Daniel José - http://www.andscacs.com