Page 1 of 1
100 easy perft(7) test positions
Posted: Fri Jul 17, 2015 12:03 pm
by sje
100 easy
perft(7) test positions
The
Perft(14) project is now tracking the easiest (i.e., minimal
perft(7) count) one hundred positions of the 96,400,068 position
unique(7) set. (Checkmate positions are excluded.) These are posted at the link
https://dl.dropboxusercontent.com/u/316 ... 4/EasyGuys whose contents will be updated regularly.
The least of the least so far is:
[d]rnbQkbnr/3ppppp/p1p5/8/8/2P5/PP1PPPPP/RNB1KBNR b KQkq - 0 4[/d]
Which has a
perft(7) count of only 67,820,026.
----
As mentioned before, the hardest 100 positions can be found at
https://dl.dropboxusercontent.com/u/316 ... /ToughGuys
Re: 100 easy perft(7) test positions
Posted: Fri Jul 17, 2015 4:19 pm
by Kotlov
How a practical use of this?
Re: 100 easy perft(7) test positions
Posted: Fri Jul 17, 2015 4:30 pm
by Auberon
Kotlov wrote:How a practical use of this?
Using Steven Edwards (OP) his amazing test cases, you can verify the correctness of your move generation.
Perft and Perft Results are briefly but well explained
here and
here respectively.
Re: 100 easy perft(7) test positions
Posted: Fri Jul 17, 2015 7:11 pm
by Kotlov
Auberon wrote:Kotlov wrote:How a practical use of this?
Using Steven Edwards (OP) his amazing test cases, you can verify the correctness of your move generation.
Perft and Perft Results are briefly but well explained
here and
here respectively.
I mean how can I use exactly these positions. What are the advantages (or disadvantages) of positions with a minimum number of nodes on other positions for the test move generation?
Re: 100 easy perft(7) test positions
Posted: Fri Jul 17, 2015 8:11 pm
by Robert Pope
I would say it's more a curiosity than anything.
And a couple of refreshed log files
Posted: Sun Jul 19, 2015 7:51 am
by sje
I have refreshed the contents of a couple of
Symbolic's log file links which were posted earlier. For easier web browser viewing, I have removed the ANSI color escape sequences.
The first is
Symbolic doing
perft(n) (n=7..10):
https://dl.dropboxusercontent.com/u/316 ... 8p9p10.log
The second is
Symbolic doing
perft(11):
https://dl.dropboxusercontent.com/u/316 ... es/p11.log
There have been some minor speed improvements for a total of about a five percent rate increase.
----
When I first calculated
perft(7) ca. 1987, it took about 36 hours; now it takes less than a second. When I first calculated
perft(10) ca. 2002, it took about two weeks; now it takes about 21 minutes.
Perft(11) now takes about 4 hours and 45 minutes.
And when running without bulk counting nor transposition table assistance,
Symbolic can run
perft() on a complex position (
BusyFEN) at the rate of about 42 million nodes per second.
----
The essentials of
Symbolic's multithreaded, semaphore-moderated job/accumulator path enumeration scheme are unchanged. That's good, because it helps prove the validity of
Symbolic's upcoming multithreaded, semaphore-moderated job/pallet α/β search scheme which should make all other searches turn
green with envy.
----
For those who want more than just a hundred or so
perft(7) test positions, here are the first ten million of them from the
unique(7) set:
https://dl.dropboxusercontent.com/u/316 ... 000.sum.gz
https://dl.dropboxusercontent.com/u/316 ... 001.sum.gz
...
https://dl.dropboxusercontent.com/u/316 ... 099.sum.gz