Internet engine matches

[BertTuyt]

Thank Michel for the precision, Kingsrow is the best program to test another. And it is clear that beat it mean that Dragon has been improved since 2.4 version. Thank for all the job that you, ED, Bert and the others doing for us, fans of software. Thank again and God bless you all!!!

Catherine thanks for your kind words.
I guess that (at least based on communicated match results) Dragon, KingsRow and Damage are in the same league.
Most likely the ELO differences (if any) are in the 5 points region.

So now the challenge is to find new ideas which will create a 30 - 50 ELO improvement.
But I'm afraid it is not easy and or straightforward.

Bert

[Krzysztof Grzelak]

Match KINGSROW - DRAGON (3-move ballots)

Kingsrow 1.54 vs. Dragon 4.4.0 (64 bit standaard) 7 wins, 6 losses , 145 draws

Kingsrow

Opening Book = Best Moves
HashTable Size = 512 MB
DB cache Size = 12000 MB, 6P
CPU 1 = 8 cores
Pondering = Off
Time = 10 Min / 65 Moves

Dragon

Opening Book = Tournament Book
HashTable Size = 512 MB
DB cache Size = 2000 MB, 6P
CPU 1 = 8 cores
Ponder= Off
Time = 10 Min / 65 Moves

[MichelG]

I guess that (at least based on communicated match results) Dragon, KingsRow and Damage are in the same league.
Most likely the ELO differences (if any) are in the 5 points region.

So now the challenge is to find new ideas which will create a 30 - 50 ELO improvement.
But I'm afraid it is not easy and or straightforward.

Bert

A 50 point improvement would be very difficult to achieve indeed. The 25w-3d-130d result Krzysztof got earlier is a massive win if it was not for the probable misconfiguration, but even that equals to 50 elo points.

The problem is the huge number of draws in this game and that is goiing to get worse with each generation of new hardware.

I still see plenty of opportunities to improve dragon in incremental ways, but in 10 years time, this game will slowly die in a pile of draws.

[Krzysztof Grzelak]

Match KINGSROW - DRAGON (3-move ballots)

Kingsrow 1.54 vs. Dragon 4.4.0 (64 bit standaard) 6 wins, 13 losses , 138 draws, 1 unknowns

Kingsrow

Opening Book = Best Moves
HashTable Size = 256 MB
DB cache Size = 12000 MB, 6P
CPU 4 = 8 cores
Pondering = Off
Time = 1 Min / 100 Moves

Dragon

Opening Book = Tournament Book
HashTable Size = 256 MB
DB cache Size = 2000 MB, 6P
CPU 4 = 8 cores
Ponder= Off
Time = 1 Min / 100 Moves

[Catherine]

Thank, Bert, we are sure that we will find the way to improve your programs, particulary, Damage, Dragon and Kingsrow; the 3 programs at the top level for me actually. good continuation.

[Krzysztof Grzelak]

Match KINGSROW - DRAGON (3-move ballots)

Kingsrow 1.54 vs. Dragon 4.4.0 (64 bit standaard) 6 wins, 4 losses, 146 draws, 2 unknowns

Kingsrow

Opening Book = Best Moves
HashTable Size = 512 MB
DB cache Size = 12000 MB, 6P
CPU 4 = 8 cores
Pondering = Off
Time = 20 Min / 75 Moves

Dragon

Opening Book = Tournament Book
HashTable Size = 512 MB
DB cache Size = 2000 MB, 6P
CPU 4 = 8 cores
Ponder= Off
Time = 20 Min / 75 Moves

[Walter]

Match KINGSROW - DRAGON (3-move ballots)

Kingsrow 1.54 vs. Dragon 4.4.0 (64 bit standaard) 6 wins, 4 losses, 146 draws, 2 unknowns

Kingsrow

Opening Book = Best Moves
HashTable Size = 512 MB
DB cache Size = 12000 MB, 6P
CPU 4 = 8 cores
Pondering = Off
Time = 20 Min / 75 Moves

Dragon

Opening Book = Tournament Book
HashTable Size = 512 MB
DB cache Size = 2000 MB, 6P
CPU 4 = 8 cores
Ponder= Off
Time = 20 Min / 75 Moves

Krzysztof, thanks for making the effort to run these matches!

Easiest conclusion is more time => more draws.
We are also getting close to all draws.
Hence, a not very risky prediction seems to me: a bit better hardware, a bit more time, a bit bigger egdb, and maybe a few search/eval optimisations => all draws.

I hope that soon more programmers will implement the Killer-rule so that we can see if that gives a new lease of life to computer draughts.

[BertTuyt]

I hope that soon more programmers will implement the Killer-rule so that we can see if that gives a new lease of life to computer draughts.

Seems an interesting exercise....
Most likely one only needs to change the MoveGenerator and re-generate again the endgame databases.
If one would stick to 6P , than I assume that with the programming efforts it could be done within a week!

I will put it anyway on my things to do list, but first I want to complete the DXP option in my Damage GUI, as several people would like to see that...

Bert

[Krzysztof Grzelak]

Match DRAGON - FLITS (2 ply balanced.pdn)

Dragon Pro 4.4.0 - Flits 11 wins, 2 losses, 145 draws

Dragon

Opening Book = Tournament Book
HashTable Size = 256 MB
DB cache Size = 2000 MB, 6P
CPU 1 = 8 cores
Ponder= Off
Time = 5 Min / 100 Moves

Flits

Transpositietabel(MB): 256 MB
Eindspel databases: all enabled
Time = 5 Min / 100 Moves

Here interesting information for Michel. Somehow strange the program is saving that's all. After 73 opening it is necessary anew to start the match.The program isn't also displaying draws.

[MichelG]

Match DRAGON - FLITS (2 ply balanced.pdn)

Dragon Pro 4.4.0 - Flits 11 wins, 2 losses, 145 draws
Here interesting information for Michel. Somehow strange the program is saving that's all. After 73 opening it is necessary anew to start the match.The program isn't also displaying draws.

Good to see this result

I don't have flits, so it is hard to tell why DXP is not working correctly with it.

[Krzysztof Grzelak]

Match DRAGON - HORIZON (2 ply balanced.pdn)

Dragon 4.4.0 - Horizon 4.3 54 wins, 104 draws

Dragon

Opening Book = Tournament Book
HashTable Size = 256 MB
DB cache Size = 2000 MB, 6P
CPU 1 = 8 cores
Ponder= Off
Time = 5 Min / 100 Moves

Horizon

Opening Book = None
HashTable Size = 256 M
DB cache Size = 2G
DB Max Piece = 6P
CPU 1 = 8 cores
Ponder= Off
Time = 5 seconds/Move

[BertTuyt]

Thanks for sharing.
So work to do for Horizon ...

Did Dragon crash during the match?
If not, than I might need to re-install Dragon and check some settings.

Bert

[BertTuyt]

Analyzing some games, I wondered if the databases were accessed.
Example game 158 move 47 3-8 loses, but with a 6P DB the draw move should be found.
I replayed this game, with 7P DB a 6P DB and no DB.
The only way I could reproduce the (wrong 3-8) move was switching off the DB.

Can you check if the DB is opened at the start?

init MoveGenerate
init Evaluation
init Done
Init HashTable, 16777216 Entries (512 MB) Allocated
init Database
status S:Databases\
DB64 Cache Init OK , 1048576 4KB Blocks Allocated
DB64 File Open = 231
DB64 ReadData Index : Start
DB64 ReadData Index : End
init Book
status S:Book\

Although directory could be different (in my case the SSD-drive S:) the DB64 File Open = .... should reflect the number of opened Databases.

In stand alone mode the DB directory should be located in the same folder as the Engine.

Bert

[Krzysztof Grzelak]

As for databases the same I don't know whether correctly he is acting. During starting programme Horizon, program is showing something like that me.

init MoveGenerate
init Evaluation
init Done
Init HashTable, 16777216 Entries (512 MB) Allocated
init Database
status Databases\
DB64 Cache Init OK , 1048576 4KB Blocks Allocated
DB64 File Open = 0
DB64 ReadData Index : Start
DB64 ReadData Index : End
init Book
status Book

I noticed that there was a serious problem in the interface Damage 2011.

[BertTuyt]

Code: Select all

DB64 File Open = 0

The DB64 File Open = 0 indicates that the program could not access the DB-Files!.
So it plays without endgame databases.

If you look to the directory which contains the Damage 2014 GUI you will see a directory Engines and Databases.
Normally when you start the Engine via the GUI it will find the Engine in the Engines directory.
Also the Engine will know the location of the Databases directory.

If you want to run a DXP match (the old GUI did not have this option) via the Engine (and you want DB-access) you just copy the Engine to the same directory as the Damage2014 GUI.
If you start now the engine from there, you should see a non-zero number for the DB64 File Open.

Could you check this, and post me the number of DB64 Files Open, so I know that all is working?

With the DB active I expect a slightly better result for Horizon but only marginally better, no dramatic improvement.

If the DB is active the Engine will also not continue playing in a DB-position.
The program assumes that all programs now have a 6P DB, and therefore in DXP-Match mode the program will adjudicate the Game automatically according to the DB-result.
In this way I also prevent some weird moves Flits sometimes makes in a DB-position.

Hope this helps.

Bert