New Version of The Baron (3.42)

[Steve Maughan]

Richard has released another version of the 3.x code base. This has a few enhancements and some additional compiles.

More info here:

http://www.chessprogramming.net/the-baron/

Steve

[tmokonen]

Good stuff. Thank you Richard, and Steve.

[CMCanavessi]

Awesome. New version will play season 2 in my tournament

[matejst]

A big thanks to Richard, and thank you, Steve, for hosting the Baron.

[Werner]

Richard has released another version of the 3.x code base. This has a few enhancements and some additional compiles.
More info here:
http://www.chessprogramming.net/the-baron/
Steve

Thanks for the new Version.
But is it really faster (not here on my core I7 7700:

The Baron
version 3.42, January 4th 2017
64-bits version
Copyright (C) 2001-2017 ir. Richard L. Pijl

Evaluation Version: evalparam.20150829184642.h

Syzygy bases enabled

CPU has POPCNT
Found 145 tablebases.

W[1]> bench 4
Doing the benchmark on depth 4
Time: 1286 ms
Time: 1470 ms
Time: 1514 ms
Time: 885 ms
Time: 2270 ms
Time: 1062 ms
Time: 599 ms
Time: 1043 ms
Time: 1278 ms
Time: 579 ms

Total Time: 23.97 seconds
Total Nodes: 29682598
Inner Nodes: 12710297 (42.8%)
Qnodes: 16292122 (54.9%)
Hash Performance: 32.17%-7.38%
QHash Performance: 6.60%-5.79%
Pawn Hash Perf: 96.95%
Check Extensions: 261050
Queen Cap Extensions: 22963
Capture Extensions: 3000
Singular Search: 3620
Singular Extensions: 45684
Cutnodes: 1918377
Allnodes: 294903
Cutnode moveordering: 92.48%
Allnode moveordering: 98.91%
False Allnodes: 11.57%
Nullmoves: 680179
Nullmove cut ratio: 38.77%
Qsearch efficiency: 27.42%
Branching factor: 2.08x
Splits: 0
Bitbase hits: 0
Syzygy WDL hits: 0
kNodes per second: 1238.22
MNodes per second: 1.21
W[1]>


The Baron
version 3.41, May 14th 2017
64-bits version
Copyright (C) 2001-2017 ir. Richard L. Pijl

Evaluation Version: evalparam.20150829184642.h

Syzygy bases enabled

CPU has POPCNT
Found 145 tablebases.

W[1]> bench 4
Doing the benchmark on depth 4
Time: 1131 ms
Time: 588 ms
Time: 636 ms
Time: 753 ms
Time: 1967 ms
Time: 959 ms
Time: 837 ms
Time: 1006 ms
Time: 868 ms
Time: 343 ms

Total Time: 18.18 seconds
Total Nodes: 24092044
Inner Nodes: 9019151 (37.4%)
Qnodes: 13207057 (54.8%)
Hash Performance: 32.34%-7.12%
QHash Performance: 8.17%-7.38%
Pawn Hash Perf: 97.06%
Check Extensions: 152308
Queen Cap Extensions: 16914
Capture Extensions: 2363
Singular Search: 0
Singular Extensions: 0
Cutnodes: 1682637
Allnodes: 188089
Cutnode moveordering: 90.51%
Allnode moveordering: 98.72%
False Allnodes: 13.21%
Nullmoves: 1865836
Nullmove cut ratio: 39.11%
Qsearch efficiency: 27.18%
Branching factor: 2.07x
Splits: 0
Bitbase hits: 0
Syzygy WDL hits: 0
kNodes per second: 1325.56
MNodes per second: 1.29
W[1]>

[Werner]

Ok,

in other testpositions the kns are higher with the new version.
It seems to me the search is not so depth.

[matejst]

Dear Richard,

I noticed that the default values in the .ini file were not the same as the values in the UCI dialogue. Which one should one use?

Then, I don't quite understand what

KingSafety (offensive) and
KingShield (defensive)

means, since I feel it should be the opposite. What do I achieve by rising these values?

Finally: do I have to activate ucioption on in the ini file for my settings in the dialogue to work?

[pijl]

Ok,

in other testpositions the kns are higher with the new version.
It seems to me the search is not so depth.

Version 3.41 was compiled with a VS 2017 professional prerelease with PGO on. After expiry of that trial license I had to revert back to VS 2017 Community that does not support PGO. So 3.42 is likely to be slower. As most changes in 3.42 are related to Syzygy probing (I changed its scoring a little, as well as changed the probing strategy) it is only useful to change to 3.42 when Syzygy are used (or when using 32-bits of course).

[pijl]

I noticed that the default values in the .ini file were not the same as the values in the UCI dialogue. Which one should one use?

The values specified in the Baron.ini will be used in the UCI dialogue as the default value. It might be that the commented out values in baron.ini do no longer reflect the hardcoded defaults (it has been some time since I last checked that). In principle: If you do not have a reason to change a value, leave it commented out.

Then, I don't quite understand what

KingSafety (offensive) and
KingShield (defensive)

means, since I feel it should be the opposite. What do I achieve by rising these values?

Increasing king safety values will increase the bonus the attacker gets for activity around the enemy king.
Increasing king shield values will increase the bonus the defender gets for its pawn shield and other defenders around its own king.
The 'bonus' could have negative value.

Finally: do I have to activate ucioption on in the ini file for my settings in the dialogue to work?

From the baron.ini file:
# ucioption (UCI): Allows setting options through UCI option commands. Setting it to off will still advertise the options,
# but the engine will ignore the commands. Values from this file will be used for all.

So ucioption should have the value on (which should also be the hardcoded default).
The reason for including this was that certain GUIs posed additional contraints on hashtable size (e.g. 32-bits versions did not allow hashtables larger than 2GB). Als the CB interface had a bug at some point where it would set the hashsize to 1MB. Making the Baron deaf for the UCI options was my way of handling these restrictions when playing in tournaments, as I could set everything I needed in the configuration file anyway.

[matejst]

Thank you, dear Richard.

It's basically what I thought, but I wanted to be sure.

I wish you the best of luck with the development of the new Baron! (hoping that you will distribute it also one day!)