Just very briefly about some other obvious SF shortcomings.
Maybe this will benefit not just SF after all.
Lessons Learned 2
Moderators: hgm, Rebel, chrisw
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Eternal knight again
[d]r2qk2r/1b3ppp/p2p1b2/2nNp3/1R2P3/2P5/1PN2PPP/3QKB1R w Kkq - 3 17
SF does not find at normal TC Rb7 above.
Suggestion how to improve SF eval for this case, as this is an eval deficiency.
Well, many engines, I am not certain about SF, also consider for outposts if the outposted piece can be attacked by enemy minor pieces, giving bigger bonus in case there are no such pieces.
This is quite OK, but it must be specified, as it turns out that the most important squares to check for such outposts are d5, e5, d6 and e6.
SF could increase substantially its knight outpost bonus (bishop outposts are really irrelevant here), only for the mg, in case an outposted knight is found on the squares d5,e5,d6 and e6 and this knight can not be attacked by enemy minors, i.e. there are enemy knights present, and no bishop the colour of the square where the knight is outposted.
You might try different values, but increase might be substantial, 1/2, 2/3 or even more. Obviously, sych a knight is bound to stay there for ever and might be worth at least a rook.
This is a subtle positional motive.
SF does not find at normal TC Rb7 above.
Suggestion how to improve SF eval for this case, as this is an eval deficiency.
Well, many engines, I am not certain about SF, also consider for outposts if the outposted piece can be attacked by enemy minor pieces, giving bigger bonus in case there are no such pieces.
This is quite OK, but it must be specified, as it turns out that the most important squares to check for such outposts are d5, e5, d6 and e6.
SF could increase substantially its knight outpost bonus (bishop outposts are really irrelevant here), only for the mg, in case an outposted knight is found on the squares d5,e5,d6 and e6 and this knight can not be attacked by enemy minors, i.e. there are enemy knights present, and no bishop the colour of the square where the knight is outposted.
You might try different values, but increase might be substantial, 1/2, 2/3 or even more. Obviously, sych a knight is bound to stay there for ever and might be worth at least a rook.
This is a subtle positional motive.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Doubled pawn specification
[d]6k1/8/2p5/2p5/2P5/8/8/6K1 w - - 0 1
SF has problems with above structures.
It needs to give additional penalty for the doubled c6 and c5 pawns, in case the more advanced of the doubled pawns is blocked by an enemy pawn, c4 above. 5-10cps additional penalty, I do not know.
What is certain is that I have watched too many games, both at LTC and blitz, where SF loses games or performs worse because of not having such a penalty.
This is a very sound chess rule and very important, so it must be present in a top engine, even if it costs something in terms of speed.
I would not be surprised if this proves to be a bigger elo winner than the patch pushed by Gary one year ago, scoring doubled pawns in terms of the distance between the two pawns.
OK, even if no one pushes such a patch in the immediate future, at least SF will be aware of the issue.
SF has problems with above structures.
It needs to give additional penalty for the doubled c6 and c5 pawns, in case the more advanced of the doubled pawns is blocked by an enemy pawn, c4 above. 5-10cps additional penalty, I do not know.
What is certain is that I have watched too many games, both at LTC and blitz, where SF loses games or performs worse because of not having such a penalty.
This is a very sound chess rule and very important, so it must be present in a top engine, even if it costs something in terms of speed.
I would not be surprised if this proves to be a bigger elo winner than the patch pushed by Gary one year ago, scoring doubled pawns in terms of the distance between the two pawns.
OK, even if no one pushes such a patch in the immediate future, at least SF will be aware of the issue.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Connected storming pawns=connected passers
Well SF scores connected pawns with the candidate passer array=somewhere half of passer array.
However, there are many connected pawns, and storming pawns are a special case, in which the bonus for being connected should be much higher, somewhere equal to the bonus for connected passers, i.e. when you score connected storming pawns, you should score the storming pawns bonus + the connected pawns bonusX2, or simply increase the value of the respective storming pawns.
I think scoring connected pawns bonusX2 or so is the right approach.
[d]6k1/5pp1/8/PP3PP1/8/8/8/6K1 w - - 0 1
the f5 and g5 storming pawns are as dangerous as the a5 and b5 passers, not less
[d]6k1/6p1/1P3pP1/P4P2/8/8/8/6K1 w - - 0 1
g6 storming pawn is not less dangerous than b6 passer
I think such a rule should be very useful for SF, as KID and related positions very frequently contain connected stroming pawns.
Quite often, SF is reluctant to connect its storming pawns in similar positions, preferring better mobility, etc. instead, but, when you show the engine a move advancing a pawn and creating connected storming pawns, the engine after a while tends to agree this might actually be the better move.
Again, storming pawns are very important, and specifying a bit more there is not a sin.
However, there are many connected pawns, and storming pawns are a special case, in which the bonus for being connected should be much higher, somewhere equal to the bonus for connected passers, i.e. when you score connected storming pawns, you should score the storming pawns bonus + the connected pawns bonusX2, or simply increase the value of the respective storming pawns.
I think scoring connected pawns bonusX2 or so is the right approach.
[d]6k1/5pp1/8/PP3PP1/8/8/8/6K1 w - - 0 1
the f5 and g5 storming pawns are as dangerous as the a5 and b5 passers, not less
[d]6k1/6p1/1P3pP1/P4P2/8/8/8/6K1 w - - 0 1
g6 storming pawn is not less dangerous than b6 passer
I think such a rule should be very useful for SF, as KID and related positions very frequently contain connected stroming pawns.
Quite often, SF is reluctant to connect its storming pawns in similar positions, preferring better mobility, etc. instead, but, when you show the engine a move advancing a pawn and creating connected storming pawns, the engine after a while tends to agree this might actually be the better move.
Again, storming pawns are very important, and specifying a bit more there is not a sin.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Re: Connected storming pawns=connected passers
Well, I am not certain that you should double the bonus, but you should increase it for sure.
Connected storming pawns comprise about 10-20% of all connected pawns, so it is about natural that you try to specify here.
Whether the increase is 10-20-30% is a matter of testing.
Connected storming pawns comprise about 10-20% of all connected pawns, so it is about natural that you try to specify here.
Whether the increase is 10-20-30% is a matter of testing.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Breakers
I am not sure this will concern SF, or maybe some other engine, what I am certain is that engines do not play this perfectly.
Please do not pay attention to the name: you should call them somehow.
A breaker is any pawn on either 4th, or 5th rank, that is open/free, i.e. the square in front of it is empty, so the pawn is mobile and able to advance further.
Breakers are due some bonus, 5cps for breakers on the 4th rank, and 10cps for breakers on the 5th rank. I think mg and eg bonus should be about the same, but it might be experimented.
It does not matter at that what the control of the square in front of the pawn is, whether own or enemy pawns and pieces control it too, the important thing is that this square is empty and the pawn is able to advance.
Why mobility for pawns is more important for pawns on the 4th and 5th ranks than pawns on other ranks?
Well, it is simply like that. On less advanced, 2nd and 3rd ranks, mobile pawns are simply not that important, as not much to achieve, while pawns on more advanced 6th and 7th ranks are already passers by definition, so a limited category, and besides such passers usually get their bonus in eval for being free.
So that, in reality, the most important mobile pawns on the board are pawn on the 4th and 5th ranks.
They are important because of at least 4 main reasons:
- pawns on the 4th and 5th ranks in the general case, when mobile, are about to gain space advantage on the 5th and 6th ranks respectively, increasing their value; a mobile pawn on the 4th rank advancing further to the 5th rank just breaks into enemy territory, and an advance of a pawn on the 5th rank confirms this advance.
- storming pawns on the 4th and 5th ranks, when mobile, are the most dangerous storming pawns, ready to break open the enemy king position. It is known fact that it is good to have blocked storming pawns on the 6th rank, but bad to have blocked storming pawns on less advanced, 4th and 5th ranks, so that mobile storming pawns on the 4th and 5th ranks are simply an asset, considerable at times
- passers on the 4th and 5th ranks are among the most valuable passers; they break into enemy territory, creating serious promotion and tactical threats. It is important for such passers to be mobile, able to advance further. Passers on the 2nd and 3rd ranks are far less dangerous, even if mobile, as not advanced, while passers on the 6th and 7th ranks are usually seen sufficiently enough in search. So that, really, the most important mobile passers with deeper positional implications that an engine search might miss are those on the 4th and 5th ranks
- in closed positions, it is precisely such mobile pawns on the 4th and 5th ranks that could potentially break open the position at an opportune time, avoiding a possible draw; so that having mobile pawns on the 4th and 5th ranks in such positions is also a valuable asset
I will now post any diagrams now, too much effort and no one is interested, but hopefully, there are some curious programmers that might try the above idea.
If you ask me, I think it is really a good idea.
Please do not pay attention to the name: you should call them somehow.
A breaker is any pawn on either 4th, or 5th rank, that is open/free, i.e. the square in front of it is empty, so the pawn is mobile and able to advance further.
Breakers are due some bonus, 5cps for breakers on the 4th rank, and 10cps for breakers on the 5th rank. I think mg and eg bonus should be about the same, but it might be experimented.
It does not matter at that what the control of the square in front of the pawn is, whether own or enemy pawns and pieces control it too, the important thing is that this square is empty and the pawn is able to advance.
Why mobility for pawns is more important for pawns on the 4th and 5th ranks than pawns on other ranks?
Well, it is simply like that. On less advanced, 2nd and 3rd ranks, mobile pawns are simply not that important, as not much to achieve, while pawns on more advanced 6th and 7th ranks are already passers by definition, so a limited category, and besides such passers usually get their bonus in eval for being free.
So that, in reality, the most important mobile pawns on the board are pawn on the 4th and 5th ranks.
They are important because of at least 4 main reasons:
- pawns on the 4th and 5th ranks in the general case, when mobile, are about to gain space advantage on the 5th and 6th ranks respectively, increasing their value; a mobile pawn on the 4th rank advancing further to the 5th rank just breaks into enemy territory, and an advance of a pawn on the 5th rank confirms this advance.
- storming pawns on the 4th and 5th ranks, when mobile, are the most dangerous storming pawns, ready to break open the enemy king position. It is known fact that it is good to have blocked storming pawns on the 6th rank, but bad to have blocked storming pawns on less advanced, 4th and 5th ranks, so that mobile storming pawns on the 4th and 5th ranks are simply an asset, considerable at times
- passers on the 4th and 5th ranks are among the most valuable passers; they break into enemy territory, creating serious promotion and tactical threats. It is important for such passers to be mobile, able to advance further. Passers on the 2nd and 3rd ranks are far less dangerous, even if mobile, as not advanced, while passers on the 6th and 7th ranks are usually seen sufficiently enough in search. So that, really, the most important mobile passers with deeper positional implications that an engine search might miss are those on the 4th and 5th ranks
- in closed positions, it is precisely such mobile pawns on the 4th and 5th ranks that could potentially break open the position at an opportune time, avoiding a possible draw; so that having mobile pawns on the 4th and 5th ranks in such positions is also a valuable asset
I will now post any diagrams now, too much effort and no one is interested, but hopefully, there are some curious programmers that might try the above idea.
If you ask me, I think it is really a good idea.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Re: Breakers
OK, it might be the case that only a mg bonus is due, so that this does not interfere in any way with an available free passer bonus in the eg.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
f6 king bind - very important!
[d]2rqnrk1/1b1nbp1p/p3p1p1/1p1pP1P1/1P1N1P2/P2B4/2P1N2P/R1BQ1RK1 w - - 0 17
SF does not see white is winning above; it thinks it is about equal, but white is winning this.
Main reason is - available f6 king bind.
I know Joona tried also with binds on f5,c5,f6 and c6 and that failed, but quite probably you should simply specify that the most important of those 4 binds to consider is of course the f6 king bind - when there are 2 white pawns on g5 and e5, binding the f6 square, with short-castled black king, on h8,g8,f8,h7,g7. But probably, the king location might simply be skipped, as kings usually castle on the king side.
So I think SF could simply give an asymmetric king bind bonus just for the f6 square, skipping the c6 square - this is very natural and the right approach IMHO.
Asymmetric terms only improve chess knowledge and tuning possibilitites, so I would not hesitate for a second to implement such an asymmetric bind.
In sharp distinction to the already existing bind for central squares, with some 16cps bonus, the f6 king bind is certainly due much more, at least 30cps SF values, but maybe well over 50cps.
In any case, it may be experimented with values. Of course, the bonus will be due just for the mg.
So that the suggestion is: white e5 and g5 pawns, then give 30cps or so bonus.
The only bonus given above for the g5 storming pawn is of course extremely insufficient.
I think SF will again utterly disregard this, although this would be very useful for SF, as it loses to many games in this way. (should I bring to your attention the Qf6 sacrifice we discussed at the beginning of this month?)
OK, does not matter, if SF is not interested, then maybe there are one or two programmers who would like to implement this in their engines - this will provide for a nice positional attacking play.
SF does not see white is winning above; it thinks it is about equal, but white is winning this.
Main reason is - available f6 king bind.
I know Joona tried also with binds on f5,c5,f6 and c6 and that failed, but quite probably you should simply specify that the most important of those 4 binds to consider is of course the f6 king bind - when there are 2 white pawns on g5 and e5, binding the f6 square, with short-castled black king, on h8,g8,f8,h7,g7. But probably, the king location might simply be skipped, as kings usually castle on the king side.
So I think SF could simply give an asymmetric king bind bonus just for the f6 square, skipping the c6 square - this is very natural and the right approach IMHO.
Asymmetric terms only improve chess knowledge and tuning possibilitites, so I would not hesitate for a second to implement such an asymmetric bind.
In sharp distinction to the already existing bind for central squares, with some 16cps bonus, the f6 king bind is certainly due much more, at least 30cps SF values, but maybe well over 50cps.
In any case, it may be experimented with values. Of course, the bonus will be due just for the mg.
So that the suggestion is: white e5 and g5 pawns, then give 30cps or so bonus.
The only bonus given above for the g5 storming pawn is of course extremely insufficient.
I think SF will again utterly disregard this, although this would be very useful for SF, as it loses to many games in this way. (should I bring to your attention the Qf6 sacrifice we discussed at the beginning of this month?)
OK, does not matter, if SF is not interested, then maybe there are one or two programmers who would like to implement this in their engines - this will provide for a nice positional attacking play.
-
jorose
- Posts: 360
- Joined: Thu Jan 22, 2015 3:21 pm
- Location: Zurich, Switzerland
- Full name: Jonathan Rosenthal
Re: f6 king bind - very important!
How is white winning this? Don't get me wrong, I'd rather have white as well, but it is not at all apparent to me that white is winning in any way. I assume your plan is something along the lines of Rf1-f3-h3 and Ne2-g3-f1-e3-g4-f6 but that is also quite slow. Am I missing anything?
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Re: f6 king bind - very important!
Hi Jonathan, we already discussed that.jorose wrote:How is white winning this? Don't get me wrong, I'd rather have white as well, but it is not at all apparent to me that white is winning in any way. I assume your plan is something along the lines of Rf1-f3-h3 and Ne2-g3-f1-e3-g4-f6 but that is also quite slow. Am I missing anything?
Please see the No chance thread:
http://talkchess.com/forum/viewtopic.ph ... 25&t=54487