Position flipping

Discussion of chess software programming and technical issues.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
sje
Posts: 4675
Joined: Mon Mar 13, 2006 6:43 pm

Position flipping

Post by sje » Mon Aug 05, 2013 6:08 am

A position flip is the action by which a position is replaced with a copy of itself modified such that each man is replaced with the corresponding man of the opposite color and then moved to the opposite side (front/back) of the board (file is same, but rank is flipped).

Other items must also be flipped: these include the color on the move, the castling status, and the en passant target square.

I notice that in Crafty, which has a flip command, there doesn't seem to be a flip of the en passant target. Am I missing something here?

Symbolic goes a bit further with its position flip. Not only are the board and the FEN scalars flipped, but also the entire move history and the saved state history. This means that it's possible to retract played moves in the flipped position all the way back to its flipped start. Also: flip(flip(P)) ≡ P

A position flip is a handy user command. It can also be used during opening book generation when (position, move) pair data are all stored using a White-to-move convention; this allows retrieval of "reversed" opening data.

mcostalba
Posts: 2684
Joined: Sat Jun 14, 2008 7:17 pm

Re: Position flipping

Post by mcostalba » Mon Aug 05, 2013 7:09 am

sje wrote:A position flip
This is another thing I am about to drop from Stockfish. The only reason while is still there is because it is from original Tord's code....but I found no use for it even once in the last years...

User avatar
michiguel
Posts: 6386
Joined: Thu Mar 09, 2006 7:30 pm
Location: Chicago, Illinois, USA
Contact:

Re: Position flipping

Post by michiguel » Mon Aug 05, 2013 7:24 am

mcostalba wrote:
sje wrote:A position flip
This is another thing I am about to drop from Stockfish. The only reason while is still there is because it is from original Tord's code....but I found no use for it even once in the last years...
It is nice for debugging purposes. In debug code, it is good to flip the position, evaluate it, and make sure the score is the same but negative. I mirror it left-right if no castle is available. This process is done every time evaluation is called (in DEBUG mode only, of course) Over the years, I caught bugs very early in this way.

Miguel

User avatar
sje
Posts: 4675
Joined: Mon Mar 13, 2006 6:43 pm

Re: Position flipping

Post by sje » Mon Aug 05, 2013 7:47 am

michiguel wrote:It is nice for debugging purposes. In debug code, it is good to flip the position, evaluate it, and make sure the score is the same but negative. I mirror it left-right if no castle is available. This process is done every time evaluation is called (in DEBUG mode only, of course) Over the years, I caught bugs very early in this way.
It can also be used during debugging for verifying that perft(P, N) = perft(flip(P), N).

mcostalba
Posts: 2684
Joined: Sat Jun 14, 2008 7:17 pm

Re: Position flipping

Post by mcostalba » Mon Aug 05, 2013 7:55 am

sje wrote:
michiguel wrote:It is nice for debugging purposes. In debug code, it is good to flip the position, evaluate it, and make sure the score is the same but negative. I mirror it left-right if no castle is available. This process is done every time evaluation is called (in DEBUG mode only, of course) Over the years, I caught bugs very early in this way.
It can also be used during debugging for verifying that perft(P, N) = perft(flip(P), N).
This IMHO should be left to an external tool that reverses FEN strings (BTW do you know one?) more than the engine itself.

User avatar
michiguel
Posts: 6386
Joined: Thu Mar 09, 2006 7:30 pm
Location: Chicago, Illinois, USA
Contact:

Re: Position flipping

Post by michiguel » Mon Aug 05, 2013 8:08 am

mcostalba wrote:
sje wrote:
michiguel wrote:It is nice for debugging purposes. In debug code, it is good to flip the position, evaluate it, and make sure the score is the same but negative. I mirror it left-right if no castle is available. This process is done every time evaluation is called (in DEBUG mode only, of course) Over the years, I caught bugs very early in this way.
It can also be used during debugging for verifying that perft(P, N) = perft(flip(P), N).
This IMHO should be left to an external tool that reverses FEN strings (BTW do you know one?) more than the engine itself.
In the case of the evaluation is not nearly as effective.

Miguel

mcostalba
Posts: 2684
Joined: Sat Jun 14, 2008 7:17 pm

Re: Position flipping

Post by mcostalba » Mon Aug 05, 2013 10:34 am

Anyhow it is easier to flip with string manipulation (at least in C++) starting from the FEN representation of a position:

Code: Select all

static char toggle_case(char c) {
  return isupper(c) ? tolower(c) : toupper(c);
}

void Position::flip() {

  string f, token;
  std::stringstream ss(fen());

  for &#40;int i = 0; i < 8; i++)
  &#123;
      std&#58;&#58;getline&#40;ss, token, i < 7 ? '/' &#58; ' ');
      std&#58;&#58;transform&#40;token.begin&#40;), token.end&#40;), token.begin&#40;), toggle_case&#41;;
      f.insert&#40;0, token + &#40;i ? "/" &#58; " "));
  &#125;

  ss >> token; // Side to move
  f += &#40;token == "w" ? "b " &#58; "w ");

  ss >> token; // Castling flags
  std&#58;&#58;transform&#40;token.begin&#40;), token.end&#40;), token.begin&#40;), toggle_case&#41;;
  f += token + " ";

  ss >> token; // En-passant square
  f += &#40;token == "-" ? token &#58; token.replace&#40;1, 1, token&#91;1&#93; == '3' ? "6" &#58; "3"));

  std&#58;&#58;getline&#40;ss, token&#41;; // Full and half moves
  f += token;

  set&#40;f, is_chess960&#40;), this_thread&#40;));

  assert&#40;pos_is_ok&#40;));
&#125;

User avatar
sje
Posts: 4675
Joined: Mon Mar 13, 2006 6:43 pm

Re: Position flipping

Post by sje » Mon Aug 05, 2013 11:04 am

michiguel wrote:In debug code, it is good to flip the position, evaluate it, and make sure the score is the same but negative. I mirror it left-right if no castle is available.
If there are no pawns on the board, you can also reflect the position about either main diagonal.

User avatar
sje
Posts: 4675
Joined: Mon Mar 13, 2006 6:43 pm

Re: Position flipping

Post by sje » Mon Aug 05, 2013 11:09 am

mcostalba wrote:Anyhow it is easier to flip with string manipulation (at least in C++) starting from the FEN representation of a position:
That code doesn't flip any position history. Move retraction will not work, and neither will repetition detection.

mcostalba
Posts: 2684
Joined: Sat Jun 14, 2008 7:17 pm

Re: Position flipping

Post by mcostalba » Mon Aug 05, 2013 11:16 am

sje wrote:
mcostalba wrote:Anyhow it is easier to flip with string manipulation (at least in C++) starting from the FEN representation of a position:
That code doesn't flip any position history. Move retraction will not work, and neither will repetition detection.
This was your previous (few hours ago) definition of flip:

Code: Select all

A position flip is the action by which a position is replaced with a copy of itself modified such that each man is replaced with the corresponding man of the opposite color and then moved to the opposite side &#40;front/back&#41; of the board &#40;file is same, but rank is flipped&#41;.

Other items must also be flipped&#58; these include the color on the move, the castling status, and the en passant target square. 
And this is also the definition most engines follow (if this makes any sense for you).

What I have written flips according to this definition.

Post Reply