Page 9 of 59

Re: 7-men Syzygy attempt

Posted: Tue Mar 27, 2018 3:56 pm
by Nordlandia
Question for Ronald: Let imagine if sf is probing 5-men + 6men adj, face identical sf probing 6-man.

How much advantage is probing 6-men during search in contrast to probing 5-men + immediate 6-men adjudication?

>--->

Another possible option is probing 5-men during search + automatic 7-man adjudication.

Re: 7-men Syzygy attempt

Posted: Tue Mar 27, 2018 4:32 pm
by guyhaw
Very impressed by your initiative - and good to see Ronald de Man active on the EGT scene again.

I would be very happy to have a version of RdM's code which created 6-man DTZ EGTs , regardless of a n-man rule.

Setting 'n' to 500 ply would I think do it!

While there is some 6-man DTZ data for pawnless endgames (c/o previous DTC work) and for 'shallow' P-ful endgames without over-deep successor endgames, we still do not have data of the quality that RdM produced in the last few years.

Here's hoping ...

g

Re: 7-men Syzygy attempt

Posted: Tue Mar 27, 2018 5:51 pm
by Jouni
What will be ELO gain of 7 piece? 5 piece gives +5 and 6 piece +3 additionally maybe. +1 ELO?

Re: 7-men Syzygy attempt

Posted: Tue Mar 27, 2018 5:52 pm
by Vinvin
syzygy wrote:My estimation was 10 TB for WDL and 10 TB for DTZ. For adjudication you would only need WDL.
Why same size for both ? WDL should be smaller, isn't it ?

Re: 7-men Syzygy attempt

Posted: Tue Mar 27, 2018 6:04 pm
by syzygy
Vinvin wrote:
syzygy wrote:My estimation was 10 TB for WDL and 10 TB for DTZ. For adjudication you would only need WDL.
Why same size for both ? WDL should be smaller, isn't it ?
The DTZ files do not include the information that can be obtained directly from the WDL files. So draws and positions with winning pawn moves and captures can be compressed out (the compressor chooses values that compress well), and win-in-10 and loss-in-10 can be stored as the same value. What also helps a lot is storing values for one side to move only.

Since there are more pawns in 7 than in 6 (so smaller DTZ on average), I expect the DTZ/WDL size ratio to be smaller (and to be less than 1). For 6 the ratio is indeed smaller than for 5.

Re: 7-men Syzygy attempt

Posted: Tue Mar 27, 2018 9:15 pm
by Dann Corbit
Jouni wrote:What will be ELO gain of 7 piece? 5 piece gives +5 and 6 piece +3 additionally maybe. +1 ELO?
It is a function of TB access speed.
If the TB is fast enough, it will be a large gain.

Re: 7-men Syzygy attempt

Posted: Wed Mar 28, 2018 7:51 am
by Ozymandias
Dann Corbit wrote:
Jouni wrote:What will be ELO gain of 7 piece? 5 piece gives +5 and 6 piece +3 additionally maybe. +1 ELO?
It is a function of TB access speed.
If the TB is fast enough, it will be a large gain.
Specially at very fast TCs.

Re: 7-men Syzygy attempt

Posted: Wed Mar 28, 2018 9:21 am
by Nordlandia
Is dtm functioning similarly to dtz when it comes to cache allocation?

Dtz only use a modest amount of cache.

Re: 7-men Syzygy attempt

Posted: Wed Mar 28, 2018 11:13 am
by syzygy
There is no reason to believe that TBs help less at long TC. I suspect that the opposite is the case.

Re: 7-men Syzygy attempt

Posted: Wed Mar 28, 2018 11:46 am
by Ozymandias
syzygy wrote:There is no reason to believe that TBs help less at long TC. I suspect that the opposite is the case.
With Houdini 4, I tested an Elo increase of at least 20 points (I don't have the exact figure at hand), and that's just going from 5-men to 6-men, not from 0 to 6. I've never see anything close to that figure, being reported.