Page 1 of 1
Who can report on BitBases efficiency in Shredder 10?
Posted: Tue Nov 21, 2006 8:04 pm
by ernest
Hi,
Shredder 9 is has been out for quite some month now, but nowhere have I seen a report on the actual gain obtained, in endgames, with the use of BitBases added to Nalimov TableBases.
I assume that the usual setting is to have both the 5-man BitBases (400 MB in memory) and the 5-man TableBases (7+ GB on disk + cache).
I personally cannot test that, because I only have 512 MB on my machine (so I didn't even buy Shredder 10
... but I have Shredder 9)
Do you have experience with Shredder 10 BitBases or have-you seen a report on their use?
(a few years ago, I experimented with Yace and 4-man BitBases, but at 4-man level, they did not seem then to bring much...)
Shredderbase contribution ...
Posted: Wed Nov 22, 2006 10:59 am
by guyhaw
Stefan M-K kindly provided me with an upgrade to SHREDDER10. The analysis window shows the counts of accesses to Shredderbases (SB) and to Endgame Tables (TB).
I am assuming, rightly or wrongly, that SB/TB is the factor of speed-up as it shows that TB rather than SB accesses to the EGTs needed to be made.
The contribution, (SB/TB)-1, increases with the scarcity of value-retaining moves. So it will be zero for KQK and KRK but maybe good for KQPKQ.
g
Re: Shredderbase contribution ...
Posted: Wed Nov 22, 2006 3:40 pm
by ernest
guyhaw wrote:Stefan M-K kindly provided me with an upgrade to SHREDDER10. The analysis window shows the counts of accesses to Shredderbases (SB) and to Endgame Tables (TB).
I am assuming, rightly or wrongly, that SB/TB is the factor of speed-up as it shows that TB rather than SB accesses to the EGTs needed to be made.
The contribution, (SB/TB)-1, increases with the scarcity of value-retaining moves. So it will be zero for KQK and KRK but maybe good for KQPKQ.
g
But concretely, has there been results reported for endgame position analysis (if not games starting with endgame suites...)?
"Position Analysis"?
Posted: Wed Nov 22, 2006 5:42 pm
by guyhaw
I'm not sure what you mean about "results for position analysis".
Sh'bases just indicate win/draw/loss and are assumed to be correct for that: they don't say whether a position is closer or further from a win. But they do filter out those moves that don't preserve current theoretical value. They are a 'speed up' rather than a better form of analysis.
g
An observation re Shredderbases
Posted: Wed Nov 22, 2006 6:19 pm
by guyhaw
I happen to be looking at a position where I see "TB 13,085 SB 1,066,774" That means that the SBs have been useful to determine that a position was won/lost (in which case, perhaps, the line could be discounted as non-optimal) or drawn (in which case the TBs do not need to be consulted).
I had not realised the 'line termination' role that a definitive 0/1 verdict might bring as a benefit. These are probably lines with a lot of captures which are not to the benefit of one or other side.
g
Re: "Position Analysis"?
Posted: Wed Nov 22, 2006 7:24 pm
by ernest
guyhaw wrote:I'm not sure what you mean about "results for position analysis".
Sh'bases just indicate win/draw/loss and are assumed to be correct for that: they don't say whether a position is closer or further from a win. But they do filter out those moves that don't preserve current theoretical value. They are a 'speed up' rather than a better form of analysis.
g
Precisely, given an end-game position (say 10 to 7-man), if you analyse with tablebases alone you get wild disk accesses, knodes/sec go down and you get THE solution (supposing there is one...) say in 24 minutes.
Now you add access to the Bitbases in RAM (Shredderbases, here
), what happens to the knodes/sec and the solution time?
Obviously this is very much position dependent!
But I was just wondering if somebody has actually experimented.