patricia devlog

[Tibono]

Hi Thorsten,

it is available in the author's Github release where the patricia-v3.exe asset actually is current 3.1 version for Windows (and without .exe should be the same for Linux).
Cheers,
Eric

[pohl4711]

Great and happy news: I'm now collaborating with Gabe, the Obsidian dev, on Patricia. He is working on helping to improve the strength of the engine so that (a) she will be able to compete with top engines, and (b) I can redouble my efforts on the style side to push the bounds of aggression even further. I am very grateful for his help and support and I am excited to see where this partnership leads. We already gained 15 elo in two days

Great news! Cant wait to see the new Patricia version...
I mentioned it before: What about transferring the neural-net of Patricia directly into the Obsidian-Engine? And continuing the development from there? Then you have the "best of both worlds", means, an already super-strong engine, with highly developed search & pruning algorithms etc. on one hand. And the incredible neural-net of Patricia on the other hand... Just my 2cents.

[pohl4711]

Where can i download a version 3.1 for pc ? I see a version for android.

I have a sub-site on my website about Patricia. There you will find all download-links. And all interesting news and my testing-results. And a bunch of spectacular wins of Patricia for replaying with the pgn-viewer.

https://www.sp-cc.de/patricia_eas_engine.htm

[Whiskers]

Great and happy news: I'm now collaborating with Gabe, the Obsidian dev, on Patricia. He is working on helping to improve the strength of the engine so that (a) she will be able to compete with top engines, and (b) I can redouble my efforts on the style side to push the bounds of aggression even further. I am very grateful for his help and support and I am excited to see where this partnership leads. We already gained 15 elo in two days

Great news! Cant wait to see the new Patricia version...
I mentioned it before: What about transferring the neural-net of Patricia directly into the Obsidian-Engine? And continuing the development from there? Then you have the "best of both worlds", means, an already super-strong engine, with highly developed search & pruning algorithms etc. on one hand. And the incredible neural-net of Patricia on the other hand... Just my 2cents.

Maybe as a variant it could work, but I want my main engine to be "Patricia", not "Patricia-obs" or "Patricia-fish" or anything like that. She is my brainchild and while I know that some people wouldn't mind if she was built off of another engine, other people (including me) would think less of the project.

[pohl4711]

Great and happy news: I'm now collaborating with Gabe, the Obsidian dev, on Patricia. He is working on helping to improve the strength of the engine so that (a) she will be able to compete with top engines, and (b) I can redouble my efforts on the style side to push the bounds of aggression even further. I am very grateful for his help and support and I am excited to see where this partnership leads. We already gained 15 elo in two days

Great news! Cant wait to see the new Patricia version...
I mentioned it before: What about transferring the neural-net of Patricia directly into the Obsidian-Engine? And continuing the development from there? Then you have the "best of both worlds", means, an already super-strong engine, with highly developed search & pruning algorithms etc. on one hand. And the incredible neural-net of Patricia on the other hand... Just my 2cents.

Maybe as a variant it could work, but I want my main engine to be "Patricia", not "Patricia-obs" or "Patricia-fish" or anything like that. She is my brainchild and while I know that some people wouldn't mind if she was built off of another engine, other people (including me) would think less of the project.

Reasonable.

[Viz]

You might as well port "aggressive net" to stockfish for what it costs, if anything stockfish is the engine when it comes to search, period.
Ofc it kinda defies the purpose of what OP tries to do.

[chesskobra]

Great news! Cant wait to see the new Patricia version...
I mentioned it before: What about transferring the neural-net of Patricia directly into the Obsidian-Engine? And continuing the development from there? Then you have the "best of both worlds", means, an already super-strong engine, with highly developed search & pruning algorithms etc. on one hand. And the incredible neural-net of Patricia on the other hand... Just my 2cents.

This is an interesting idea. So I compiled Obsidian 13.0 with the net that comes with Patricia 3.1 (abby.nnue). Obsidian 13.0 (Abby) lost two games to Patricia 3.1, each in under 35 moves. Could there be something wrong with my compile? I just changed the eval file to "abby.nnue" in the nnue.h file. That is the only file in which original eval file "net45.bin" is mentioned.

On the other hand Obsidian 13.0 with the original net (net45.bin) (again my own compile on linux) won two games against Patricia 3.1 without much difficulty, which is expected.

A general question: which top engines have interchangeable nets (i.e., one engine can be compiled with a net from the other)?

[mclane]

Where can i download a version 3.1 for pc ? I see a version for android.

I have a sub-site on my website about Patricia. There you will find all download-links. And all interesting news and my testing-results. And a bunch of spectacular wins of Patricia for replaying with the pgn-viewer.

https://www.sp-cc.de/patricia_eas_engine.htm

Very good website !! Thank you.

[Whiskers]

The dev version of Patricia is now 100 ELO stronger than Patricia 3.1. Collaborating with the Obsidian dev has been busy but very fruitful, he just threw himself at working on it lol. I really appreciate his help!

I now have 1.6 billion positions of Patricia data. Almost there...

[Whiskers]

First datagen run was a disaster.
oops.
Currently figuring out what I broke in between testing and actually generating the large dataset.