penalty for lack of growth

[lech]

Users of engines („go infinite ") have a problem with the fortress, „crazy piece” or magnetic passed pawn.
I think this can easily and safely be solved by a penalty for lack of growth (e.g. UCI option).

Algorithm:

Code: Select all

depth: x
scores: x’, x”
penalty: p
expected increase: y
threshold (positive): w 

on x depth a move gets score x’ 
on (x + 1) depth the same move gets score x”

if ((x” > w)  & (abs(x” – x’) < y) )
{i= i + 1 + (x” < = x’);
  x” = x’; // to break it (if stable small increase) or to avoid a swing
  score of this move =  Max(0, x” – p * i); }
else {i=0;
 score of this move =  x”;}
 

[Dann Corbit]

Users of engines („go infinite ") have a problem with the fortress, „crazy piece” or magnetic passed pawn.
I think this can easily and safely be solved by a penalty for lack of growth (e.g. UCI option).

Algorithm:

Code: Select all

depth: x
scores: x’, x”
penalty: p
expected increase: y
threshold (positive): w 

on x depth a move gets score x’ 
on (x + 1) depth the same move gets score x”

if ((x” > w)  & (abs(x” – x’) < y) )
{i= i + 1 + (x” < = x’);
  x” = x’; // to break it (if stable small increase) or to avoid a swing
  score of this move =  Max(0, x” – p * i); }
else {i=0;
 score of this move =  x”;}
 

What if the lack of growth move is the only move that saves a draw?
We might penalize it and switch to a losing move.

[lech]

Users of engines („go infinite ") have a problem with the fortress, „crazy piece” or magnetic passed pawn.
I think this can easily and safely be solved by a penalty for lack of growth (e.g. UCI option).

Algorithm:

Code: Select all

depth: x
scores: x’, x”
penalty: p
expected increase: y
threshold (positive): w 

on x depth a move gets score x’ 
on (x + 1) depth the same move gets score x”

if ((x” > w)  & (abs(x” – x’) < y) )
{i= i + 1 + (x” < = x’);
  x” = x’; // to break it (if stable small increase) or to avoid a swing
  score of this move =  Max(0, x” – p * i); }
else {i=0;
 score of this move =  x”;}
 

What if the lack of growth move is the only move that saves a draw?
We might penalize it and switch to a losing move.

It is for „go infinite” and a mindful choice (by UCI option).
This move on the end of analyze gets 0 score. If the other can get higher score, it is the better move.
It will be welcome by many of us
Is it better to see a bad score in the fortress, „crazy piece” or magnetic passed pawn and many endgames????

[lech]

I corrected the algorithm. It is interesting that it can solve the “null move problem” too.
How to set values x, y, w, m?
If the values are set correctly it is impossible to lose a good result.

Code: Select all

depth: x 
scores: x1, x2 
penalty: p 
expected increase: y 
threshold : w 
min score: m

on x depth a move gets score x1 
on (x + 1) depth the same move gets score x2 

if ((x2 > w)  && (abs(x2 – x1) < y) ) 
{i = i + 1 + (x2 < = x1); 
x2  = x1;  // to break if is a stable increase
 score of this move =  Max(m, x2 – p * i); } 
else {i=0; 
 score of this move =  x2}

[lech]

Thanks for applause!
I understand that the immediate Dann Corbit’s (well-thought-out ! ) opinion was bought by others.
My dedication, for all creators of ELO-engines (clones too ) so-called “chess engines”, is the following position:
[d] 8/5pKN/5Pp1/5kP1/1B6/8/b7/7q b - -
If your engine is able to make (“go infinite”) 1...Qxh7 simple move, you may use the name “chess engine” for your product.

[marcelk]

Thanks for applause! :o :lol:
I understand that the immediate Dann Corbit’s (well-thought-out ! 8-) ) opinion was bought by others. :cry:
My dedication, for all creators of ELO-engines (clones too :wink: ) so-called “chess engines”, is the following position:
[d] 8/5pKN/5Pp1/5kP1/1B6/8/b7/7q b - -
If your engine is able to make (“go infinite”) 1...Qxh7 simple move, you may use the name “chess engine” for your product. :lol:

Marek,

1. Is this a position from an actual game or is it a study?

2. Did you test your lack-of-growth-algorithm in this position? Did it find Qxh7? Can you show the engine output?

[lech]

1. Is this a position from an actual game or is it a study?
2. Did you test your lack-of-growth-algorithm in this position? Did it find Qxh7? Can you show the engine output?

Is it a strange position ?
I am preparing a GUI (for myself) with option:
MuliPV = all possible moves.

I will test my algorithm with e.g. setting:
w = 100 (for SF)
y = (x2 – w) / 40
p = y
m = w + y // „+ y” if there is no a better hidden move

Code: Select all

 depth: x 
scores: x1, x2 
penalty: p 
expected increase: y 
threshold : w 
min score: m 

on x depth a move gets score x1 
on (x + 1) depth the same move gets score x2 

if ((x2 > w)  && (abs(x2 – x1) < y) ) 
{i = i + 1 + (x2 < = x1); 
x2  = x1;  // to break if is a stable increase 
 score of this move =  Max(m, x2 – p * i); } 
else {i=0; 
 score of this move =  x2}

I believe it should work correctly. Of course, only if a hidden better move (like 1...Qxh7) is one of first moves.
[d] 8/1q3pKN/5Pp1/5kP1/1B6/8/b7/8 b - -
The solution 1...Qh1 and 2...Qxh7 can be find only after some changes in codes of engines.