Engin wrote: ↑Fri Dec 25, 2020 7:39 pm
At the moment i am very pissed off
There are some trolls around that complains everything what programmers are doing, even they are spend many years at own work and let the community enjoy for free !
Hi Engin,
Your anger is very understandable.
I have another implementation that is not GPL but currently it is not compatible with SF nets. https://github.com/dshawul/nncpu-probe
It uses vertical symmetry to reduce the number of input nodes by half. What I suggest is you simply keep
on training your own net. By that time either I would finish support for SF format nets or you would have implemented
your own nnue prober anyway.
Daniel
Wow !
great news Daniel
Can i use and training my own net with it and is it allowed under the GPL to use ??
Currently i am using EPD for training my own simple network (64, 32, 1) where the scores are allways from white view.
if your nn code is also possible to train the network with epd positions then its really nice to have it. i mean not only forward probe the network, it should be also possible to train with backward propagation to change the weights and save/load the net too.
Last edited by Engin on Fri Dec 25, 2020 8:53 pm, edited 5 times in total.
Damir wrote: ↑Fri Dec 25, 2020 5:38 pm
As long as we get Tornado for free i don't see any harm in using ideas from other engines. It would have been completely different if Tornado was commercial engine. I think Madeleine better think before she start accusing people of stealing other engine's code.
I am not accusing anybody of stealing anything. Engin specifically stated on the Tornado website that Tornado was using the nnue-probe from Daniel Shawul, which uses the GPLv3 license.
There has been some discussion recently regarding why the TCEC administrators aren't allowing Tornado to participate in TCEC, and one of the things they brought up was the GPLv3 violation.
i don't care about TCEC what say are doing, pissed off from them too
Engin wrote: ↑Fri Dec 25, 2020 8:29 pm
Can i use and training my own net with it and is it allowed under the GPL to use ??
For your information, Daniel Shawul's license for nncpu-probe isn't the GPL, it is a different license much closer to public domain than the GPL is. Unlike the GPL, it doesn't require you to make your code open source, so you could keep your code closed source and only provide the binary, just as you are doing now.
Engin wrote: ↑Fri Dec 25, 2020 8:29 pm
Can i use and training my own net with it and is it allowed under the GPL to use ??
For your information, Daniel Shawul's license for nncpu-probe isn't the GPL, it is a different license much closer to public domain than the GPL is. Unlike the GPL, it doesn't require you to make your code open source, so you could keep your code closed source and only provide the binary, just as you are doing now.
If you are using GPL'ed code, you are obliged to fulfill any source request from a party whom you've shared a binary. If you post Tornado on your website, every one who has downloaded it now has a right to request the source from you. If your program is able to be fully separated from NNUE, you can skirt this and share only the sections with NNUE. However, if your program is not able to fully function without the GPL'ed code, which it likely is, then you have that obligation still.
If you don't want to share your code, you have the ability to send a binary only to TCEC. TCEC then has the right to request the source, but its possible that they will not, and if they did then they would likely do it in confidence. That is up to them, and you could figure that out with them privately.
If you are mad that people like me think that your copied NNUE is boring, get over it. Either reject the notation and ignore people in my camp, or make some headway into changing our minds by implementing it on your own. Changing the architecture. Coming up with your own trainer and training solutions. Again, if you want to just drag-and-drop the NNUE in, that is your prerogative. Just don't expect everyone to be fully on board. Although it seems all the rating lists and tournaments are in support.
No one is stopping you from working on Tornado for your own joy. If you share Tornado, you have to abide by the licenses of the works you have used. Anything less is a violation of the letter and the spirit of the license.
#WeAreAllDraude #JusticeForDraude #RememberDraude #LeptirBigUltra "Those who can't do, clone instead" - Eduard ( A real life friend, not this forum's Eduard )
AndrewGrant wrote: ↑Fri Dec 25, 2020 8:58 pm
If you are using GPL'ed code, you are obliged to fulfill any source request from a party whom you've shared a binary. If you post Tornado on your website, every one who has downloaded it now has a right to request the source from you. If your program is able to be fully separated from NNUE, you can skirt this and share only the sections with NNUE. However, if your program is not able to fully function without the GPL'ed code, which it likely is, then you have that obligation still.
If you don't want to share your code, you have the ability to send a binary only to TCEC. TCEC then has the right to request the source, but its possible that they will not, and if they did then they would likely do it in confidence. That is up to them, and you could figure that out with them privately.
If you are mad that people like me think that your copied NNUE is boring, get over it. Either reject the notation and ignore people in my camp, or make some headway into changing our minds by implementing it on your own. Changing the architecture. Coming up with your own trainer and training solutions. Again, if you want to just drag-and-drop the NNUE in, that is your prerogative. Just don't expect everyone to be fully on board. Although it seems all the rating lists and tournaments are in support.
No one is stopping you from working on Tornado for your own joy. If you share Tornado, you have to abide by the licenses of the works you have used. Anything less is a violation of the letter and the spirit of the license.
many thanks to de-motivate me again from chess programming !
Engin wrote: ↑Fri Dec 25, 2020 8:55 pm
after that you complain even this maybe
Maybe I didn't word it clearly enough. I didn't mean it as a complaint; nncpu-probe not being GPLv3.0 is a good thing for you, because it allows you to use Daniel Shawul's nncpu-probe code without having to release your source code along with your compiled binary. This is different from nnue-probe, which requires you to release your source code along with your compiled binary.
AndrewGrant wrote: ↑Fri Dec 25, 2020 8:58 pm
If you are using GPL'ed code, you are obliged to fulfill any source request from a party whom you've shared a binary. If you post Tornado on your website, every one who has downloaded it now has a right to request the source from you. If your program is able to be fully separated from NNUE, you can skirt this and share only the sections with NNUE. However, if your program is not able to fully function without the GPL'ed code, which it likely is, then you have that obligation still.
If you don't want to share your code, you have the ability to send a binary only to TCEC. TCEC then has the right to request the source, but its possible that they will not, and if they did then they would likely do it in confidence. That is up to them, and you could figure that out with them privately.
If you are mad that people like me think that your copied NNUE is boring, get over it. Either reject the notation and ignore people in my camp, or make some headway into changing our minds by implementing it on your own. Changing the architecture. Coming up with your own trainer and training solutions. Again, if you want to just drag-and-drop the NNUE in, that is your prerogative. Just don't expect everyone to be fully on board. Although it seems all the rating lists and tournaments are in support.
No one is stopping you from working on Tornado for your own joy. If you share Tornado, you have to abide by the licenses of the works you have used. Anything less is a violation of the letter and the spirit of the license.
many thanks to de-motivate me again from chess programming !
merry christmas!
Quite the opposite.
No one is stopping you from working on Tornado for your own joy.
You can do whatever you find joy in. If you share with the world, people will hold you to follow the rights of others. The rules are very fair. Most engines now are open source. You likely don't have any super secret elo winning ideas that have not been considered. I don't even think Komodo has those.
#WeAreAllDraude #JusticeForDraude #RememberDraude #LeptirBigUltra "Those who can't do, clone instead" - Eduard ( A real life friend, not this forum's Eduard )
Engin wrote: ↑Fri Dec 25, 2020 7:39 pm
At the moment i am very pissed off
There are some trolls around that complains everything what programmers are doing, even they are spend many years at own work and let the community enjoy for free !
Hi Engin,
Your anger is very understandable.
I have another implementation that is not GPL but currently it is not compatible with SF nets. https://github.com/dshawul/nncpu-probe
It uses vertical symmetry to reduce the number of input nodes by half. What I suggest is you simply keep
on training your own net. By that time either I would finish support for SF format nets or you would have implemented
your own nnue prober anyway.
Daniel
i never understand why the input must be so big, i am using just 64 input neurons (for pieces from 1 - 6, black has negative ones) for the whole board and just only one hidden layer with 32 neurons, the output has only 1 neuron that output from -1 to +1 score from white view of point. that is enough and should be very fast calculate on the cpu
if your nn_probe is only probing so how i can training it without a function that can backpropagate the network ??
