Rival Chess 1.0.1 and 1.0.3

[tmokonen]

I was on GitHub today, and I noticed that Chris Moreton had an updated version of his engine Rival on that site. The latest version I had been aware of previously was version 0094. The repository is located at:

https://github.com/chris-moreton/rival- ... oid-engine

The description of the program states "The Rival chess Java engine as used in the Rival Chess Android App." However, if you download either the 1.0.1 or 1.0.3 release, there will be a jar file named RivalChess.jar in the dist folder, which can be run as a UCI engine in Windows, or, presumably, any other platform capable of running Java.

I tried both the 1.0.1 and 1.0.3 versions, and there seems to be a bug in the 1.0.3 version. It works fine if you do a fixed depth or fixed time search, but it will look only one ply deep if using a time control other than fixed time. 1.0.1 runs OK with any time control. I tried these versions by using a batch file within Arena. If anyone can get 1.0.3 running properly with a non-fixed time control. let me know.

The author states that the latest version is about 200-300 ELO stronger than Rival 1.1.8, which is rated 1976 on the CCRL 40/40 rating list.

[Guenther]

I was on GitHub today, and I noticed that Chris Moreton had an updated version of his engine Rival on that site. The latest version I had been aware of previously was version 0094. The repository is located at:

https://github.com/chris-moreton/rival- ... oid-engine

The description of the program states "The Rival chess Java engine as used in the Rival Chess Android App." However, if you download either the 1.0.1 or 1.0.3 release, there will be a jar file named RivalChess.jar in the dist folder, which can be run as a UCI engine in Windows, or, presumably, any other platform capable of running Java.

I tried both the 1.0.1 and 1.0.3 versions, and there seems to be a bug in the 1.0.3 version. It works fine if you do a fixed depth or fixed time search, but it will look only one ply deep if using a time control other than fixed time. 1.0.1 runs OK with any time control. I tried these versions by using a batch file within Arena. If anyone can get 1.0.3 running properly with a non-fixed time control. let me know.

The author states that the latest version is about 200-300 ELO stronger than Rival 1.1.8, which is rated 1976 on the CCRL 40/40 rating list.

Thanks for the info Tony. I had RivalJava 1.01 already in the chronology,
because it was already released in 2015 in a SJCE package, but I missed
the github link, thus I couldn't watch it.

I will try 1.03 later today. It is around 10 times the size of 1.01?

Guenther

[tmokonen]

Thanks for the info Tony. I had RivalJava 1.01 already in the chronology,
because it was already released in 2015 in a SJCE package, but I missed
the github link, thus I couldn't watch it.

I will try 1.03 later today. It is around 10 times the size of 1.01?

Guenther

That's interesting, because the latest version of SJCE from the SourceForge site (23-03-17) has the older Rival 0094. I wonder why the newer version was taken out and replaced with the older version.

I am looking at Chris's commits, at

https://github.com/chris-moreton/rival- ... its/master

On Sep 21, 2016, he has a commit to Add JAR, which is a much smaller JAR file, then a second commit titled Update JAR to be runnable, which is where the size increase occurs.

[Guenther]

Thanks for the info Tony. I had RivalJava 1.01 already in the chronology,
because it was already released in 2015 in a SJCE package, but I missed
the github link, thus I couldn't watch it.

I will try 1.03 later today. It is around 10 times the size of 1.01?

Guenther

That's interesting, because the latest version of SJCE from the SourceForge site (23-03-17) has the older Rival 0094. I wonder why the newer version was taken out and replaced with the older version.

I am looking at Chris's commits, at

https://github.com/chris-moreton/rival- ... its/master

On Sep 21, 2016, he has a commit to Add JAR, which is a much smaller JAR file, then a second commit titled Update JAR to be runnable, which is where the size increase occurs.

I see he added an uci nodes limit option between both jars - may be that is always active - just a guess?

I also downloaded the 09/2016 version of 1.03 now and I bet it will work.
Testing in a few minutes...

[Guenther]

Thanks for the info Tony. I had RivalJava 1.01 already in the chronology,
because it was already released in 2015 in a SJCE package, but I missed
the github link, thus I couldn't watch it.

I will try 1.03 later today. It is around 10 times the size of 1.01?

Guenther

That's interesting, because the latest version of SJCE from the SourceForge site (23-03-17) has the older Rival 0094. I wonder why the newer version was taken out and replaced with the older version.

I am looking at Chris's commits, at

https://github.com/chris-moreton/rival- ... its/master

On Sep 21, 2016, he has a commit to Add JAR, which is a much smaller JAR file, then a second commit titled Update JAR to be runnable, which is where the size increase occurs.

I see he added an uci nodes limit option between both jars - may be that is always active - just a guess?

I also downloaded the 09/2016 version of 1.03 now and I bet it will work.
Testing in a few minutes...

Well, the earlier 1.03 version does not run at all and immediately crashes.
The later one with the big size only 'searches' 2 nodes for depth 1 and it always plays the same.
And most important there is no uci option to set the nodes # at all.

At least this way it can fill the gap between RandMovers and AndWorst in experiments ;-)

Guenther

[tmokonen]

Well, the earlier 1.03 version does not run at all and immediately crashes.
The later one with the big size only 'searches' 2 nodes for depth 1 and it always plays the same.
And most important there is no uci option to set the nodes # at all.

At least this way it can fill the gap between RandMovers and AndWorst in experiments

Guenther

You can search by fixed nodes (UCI go nodes command), and that will work. If you then select a time control afterward, it will still search by number of fixed nodes. It looks like adding this feature broke the time handling.

[Guenther]

I tried v1.0.1 under Arena and while it was playing task manager showed permanent CPU usage. I don't know if that means pondering or just a background process used for something. There were no immediate responses to the opponent's moves.

Ouch you are right and even worse it uses up to 2 cpus when on move.
I verified this also for the previously known version 0.094.
It seems it uses one full cpu for permanent polling or similar...