Perft(14) Weekly Status Reports for 2016

[sje]

Castling does not reset the half move counter. Only captures and pawn moves do this.

[Dann Corbit]

Castling does not reset the half move counter. Only captures and pawn moves do this.

True but castling is still not a reversible move.
I guess it goes into the same quiet category as moving a rook or king which formerly had castle rights

[sje]

Thank you for the verification; every calculation helps.

I do not understand the cause for the differences. Could it be an over-heated video card? I have seen this happen in Apple notebooks and desktops, including relatively recent iMac machines.

I will wait until Ankan has a chance to take a further look before I check the other differences.

----

Symbolic has produced more than 55,800,000 perft(7) results so far, about 57.88% of the 96,400,068 needed.

[sje]

I suppose that detecting and recording those metrics could help debug a chess program. But otherwise they won't help much with move selection, so if they're included in a program, then they should be conditionally compiled so as to not slow the program during regular play.

----

For those wanting to test a program using a big bunch of checkmate and stalemate positions, try the files at: https://drive.google.com/drive/folders/ ... zFibVJicms

Each file has some 100,000 FEN positions. How fast can YOUR program run through these files?

[Dann Corbit]

I suppose that detecting and recording those metrics could help debug a chess program. But otherwise they won't help much with move selection, so if they're included in a program, then they should be conditionally compiled so as to not slow the program during regular play.
{snip}

I am interested in the mathematical patterns.
What is the growth rate of checks?
What is the growth rate of checkmates?
What is the growth rate of quiet moves?
What is the growth rate pf captures?
What is the exact branching factor of pure mini-max for each ply?
The last question is one of the questions answered by the simple move counting. But what about all the other questions?
Instead of one number, I am interested in many numbers. And especially to see if there are patterns.

It will be intensely interesting if we can see a clear, mathematical pattern in these things and (for instance) predict the results for future plies.

[sje]

Perft(14) Weekly Status 2016-09-04

Symbolic has produced more than 56,100,000 perft(7) results so far, about 58.19% of the 96,400,068 needed.

Day count: 760
Estimated remaining day count: 546
Estimated total day count: 1,306

Average throughput: 73,816 results/day
Effective frequency: 238.06 GHz

Work units not yet started (390): 574-963

Sum of perft()s: 1,279,258,909,651,533,598
Sum of products: 15,579,797,542,236,567,395