Will TCEC need to rewrite their rules?

[dkappe]

Just to be clear, I have no issues with lc0 or Allie. My issue is with the somewhat arbitrary and vague TCEC rules governing Neural Network engine participation.

These rules state:

Definition
A neural network is a computer system modeled on the human brain and nervous system. For the purpose of TCEC a participant is considered a neural network (NN) engine if it generally requires the use of GPU and consists of at least the following 3 parts:
1. The code for training the neural network
2. The neural network (and weights file) itself
3. The engine that executes this network

It is the parts 2 and 3 that will actually be a playing combination at TCEC. Part 1 is used in preparation.

Uniqueness
For an NN engine to be unique in the TCEC context, at least two of the three defining parts mentioned above have to be unique.

We know both from the author of Allie and a leading lc0 dev that Allie uses a (rather important) part of lc0 code. Also it seems that the net in question used some lc0 training data and the training code.

One could argue that Allie + Stein qualifies in zero of the three areas, unless of course “unique” means whatever you want it to mean.

I would feel more comfortable if TCEC were to clarify their rules.

[gonzochess75]

Not interested in using this thread to defend Allie, but want to point out that any future rules you might come up with should account for the fact that Ethereal, Arasan, and Lc0 (probably a lot more?) are all using mostly the exact same (very important) code that was originally ported from Stockfish for tablebase support.

Inspite is this, I think it unreasonable to suggest these engines are not “unique” even though they share some pretty important code.

Also, might need to account for fact that proprietary engines might well use the same code (and maybe a lot more) that is released under a more liberal open source license.

[Ratosh]

TB support is a feature, all those engines work without this feature. I think TCEC is right demanding uniqueness but needs to be clear about uniqueness and fair use.

[AndrewGrant]

Just to be clear, I have no issues with lc0 or Allie. My issue is with the somewhat arbitrary and vague TCEC rules governing Neural Network engine participation.

These rules state:

Definition
A neural network is a computer system modeled on the human brain and nervous system. For the purpose of TCEC a participant is considered a neural network (NN) engine if it generally requires the use of GPU and consists of at least the following 3 parts:
1. The code for training the neural network
2. The neural network (and weights file) itself
3. The engine that executes this network

It is the parts 2 and 3 that will actually be a playing combination at TCEC. Part 1 is used in preparation.

Uniqueness
For an NN engine to be unique in the TCEC context, at least two of the three defining parts mentioned above have to be unique.

We know both from the author of Allie and a leading lc0 dev that Allie uses a (rather important) part of lc0 code. Also it seems that the net in question used some lc0 training data and the training code.

One could argue that Allie + Stein qualifies in zero of the three areas, unless of course “unique” means whatever you want it to mean.

I would feel more comfortable if TCEC were to clarify their rules.

This seems more like a contrived reason to push DeusX as a non-clone, which of course it is not.

[hgm]

Well, they won't have to do anything, because they are not a serious championship, accountable to no one, and can do whatever they like. Obviously they think there is some entertainment value in having several, differently tuned clones on a NN engine in their tournament. In the long run, when many truly different NN engines will exist, this will become an annoyance rather than an asset, and they will adapt the rules accordingly.

[syzygy]

Not interested in using this thread to defend Allie, but want to point out that any future rules you might come up with should account for the fact that Ethereal, Arasan, and Lc0 (probably a lot more?) are all using mostly the exact same (very important) code that was originally ported from Stockfish for tablebase support.

That code is not coming originally from Stockfish.

[Jesse Gersenson]

It comes from Cfish.

[pilgrimdan]

Just to be clear, I have no issues with lc0 or Allie. My issue is with the somewhat arbitrary and vague TCEC rules governing Neural Network engine participation.

These rules state:

Definition
A neural network is a computer system modeled on the human brain and nervous system. For the purpose of TCEC a participant is considered a neural network (NN) engine if it generally requires the use of GPU and consists of at least the following 3 parts:
1. The code for training the neural network
2. The neural network (and weights file) itself
3. The engine that executes this network

It is the parts 2 and 3 that will actually be a playing combination at TCEC. Part 1 is used in preparation.

Uniqueness
For an NN engine to be unique in the TCEC context, at least two of the three defining parts mentioned above have to be unique.

We know both from the author of Allie and a leading lc0 dev that Allie uses a (rather important) part of lc0 code. Also it seems that the net in question used some lc0 training data and the training code.

One could argue that Allie + Stein qualifies in zero of the three areas, unless of course “unique” means whatever you want it to mean.

I would feel more comfortable if TCEC were to clarify their rules.

Rybka has opened a can of worms that is still annoying everyone to this day … too bad … it did not have to be this way … if this is what Vas wanted … then he has done a good job … I would imagine Vas is pretty pleased with himself right now …

[mwyoung]

Just to be clear, I have no issues with lc0 or Allie. My issue is with the somewhat arbitrary and vague TCEC rules governing Neural Network engine participation.

These rules state:

Definition
A neural network is a computer system modeled on the human brain and nervous system. For the purpose of TCEC a participant is considered a neural network (NN) engine if it generally requires the use of GPU and consists of at least the following 3 parts:
1. The code for training the neural network
2. The neural network (and weights file) itself
3. The engine that executes this network

It is the parts 2 and 3 that will actually be a playing combination at TCEC. Part 1 is used in preparation.

Uniqueness
For an NN engine to be unique in the TCEC context, at least two of the three defining parts mentioned above have to be unique.

We know both from the author of Allie and a leading lc0 dev that Allie uses a (rather important) part of lc0 code. Also it seems that the net in question used some lc0 training data and the training code.

One could argue that Allie + Stein qualifies in zero of the three areas, unless of course “unique” means whatever you want it to mean.

I would feel more comfortable if TCEC were to clarify their rules.

Rybka has opened a can of worms that is still annoying everyone to this day … too bad … it did not have to be this way … if this is what Vas wanted … then he has done a good job … I would imagine Vas is pretty pleased with himself right now …

TCEC will and can change any rules they wish. To allow any program to play. If TCEC thinks this is in their best interest for a program to play. TCEC is a computer exhibition. Not any kind of real computer chess championship.

FAQ
Is this the official World Computer Chess Championship?
No, it is not, although some people regard it as such.

The WCCC originality rule.


TOURNAMENT RULE 2 (Originality rule)

Each program must be the original work of the entering developers, possibly with the inclusion of game playing code and/or data from other sources for which the entering developers have a legal right of use. Developers whose code is derived from or includes (1) game-playing code; and/or (2) data written by others, must name (a) all the other developers of whom they are aware; and (b) the source of such code and/or data, in their tournament registration details

[MikeB]

Not interested in using this thread to defend Allie, but want to point out that any future rules you might come up with should account for the fact that Ethereal, Arasan, and Lc0 (probably a lot more?) are all using mostly the exact same (very important) code that was originally ported from Stockfish for tablebase support.

That code is not coming originally from Stockfish.

+1 ... how quickly people forget ...