Programming Puzzles, Chess Positions and Huffman Coding | Hacker News Reader

Programming Puzzles, Chess Positions and Huffman Coding

cforcoding.com

15 pointsfogus6 comments

Arithmetic encoding is usually more space-efficient than Huffman encoding, but it's not used frequently because it has much higher computational complexity for very small gain. However, if you are looking for the most space-efficient representation, it's worth taking a look at.

http://en.wikipedia.org/wiki/Arithmetic_encoding

Arithmetic compression differs from Huffman encoding because it encodes the entire dataset as a whole, rather than symbol-by-symbol. At worst, it will provide the same encoding that Huffman does, and it has far better performance on degenerate cases (such as 1 symbol having 75% chance). I think the commonly cited statistic is about 5% better on average, for some definition of 'average.'

For the curious: Arithmetic encoding works by taking all possible sequences, and taking the closed interval from 0 to 1 and chopping it into pieces, where the size of each piece is the probability of the corresponding sequence. Then the encoding of a sequence is the decimal (well, binary) expansion of any number that falls within the corresponding interval.

Semi-related - for those who are interested in writing their own chess programs - the de facto standard method of representing positions (in a human readable form) is to use http://en.wikipedia.org/wiki/Forsyth-Edwards_Notation (FEN).

The starting board position in FEN would look something like this: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1

I'm surprised http://en.wikipedia.org/wiki/Zobrist_hashing wasn't mentioned at all in the article.

To be honest with you I'd never heard of Zobrist hashing but then again I've never tried to write a Chess or other such board game for which that it is appropriate for decision trees and such.

I'll stress that the idea of this post isn't to implement a Chess engine but merely to examine an admittedly artificial problem (much like Code Golf results in code you usually wouldn't actually use for anything real because it's typically unreadable even if clever).

Various hashing techniques could speed up comparisons of positions but they don't allow you to store positions.

A nice article about how to encode the state of a chess game at any point which recognizes that Morse code is a form of and in some small way anticipates Huffman encoding.

En passant only needs 3 bits to encode the file ("column") containing the eligible pawn. The rank ("row") is inferred from the context -- whose move it is. (W=5, B=4)

3 bits always suffice to indicate either (a) a pawn en prise (as it were) or (b) no en passant possible because there's no enemy pawn on the proper square for that file. (You can't have pawns on all 8 files and at the same time have one eligible for en passant capture, since you've got to have an opponent's pawn on an adjacent file. So in the usual case of no en passant you just pick a file not occupied by a pawn of the right color, alternatively one of the right color not adjacent to a pawn of the color to move. This is always possible).

Reply on news.ycombinator.com