CCRL Discussion Board

eboatwright wrote: ↑Sun Jan 28, 2024 4:25 pm And a new patch has been put out
https://github.com/eboatwright/Maxwell/ ... g/v3.0.8-2

Although if you have to add this as separate version, you an probably just skip it, because I think v3.1 might be out soon

Thank you!

I will skip it. And I suggest you take your time with the next version, it is not likely to be tested immediately as my testing queue is quite long. Although I am not the only tester, other testers do not seem particularly interested in these medium-strength engines.

Gabor Szots wrote: ↑Sun Jan 28, 2024 5:40 pm

eboatwright wrote: ↑Sun Jan 28, 2024 4:25 pm And a new patch has been put out
https://github.com/eboatwright/Maxwell/ ... g/v3.0.8-2

Although if you have to add this as separate version, you an probably just skip it, because I think v3.1 might be out soon

Thank you!

I will skip it. And I suggest you take your time with the next version, it is not likely to be tested immediately as my testing queue is quite long. Although I am not the only tester, other testers do not seem particularly interested in these medium-strength engines.

Ok thanks, good to know

Maxwell v3.1 is out!!
https://github.com/eboatwright/Maxwell/ ... s/tag/v3.1

Test results vs v3.0.8-2 I'm real excited to see this one's rating

There's gotta be a running joke that you only find bugs after you make a major release, and here it is
This patch fixes a bug where the engine would crash if the Hash size was set to 0 MB

https://github.com/eboatwright/Maxwell/ ... tag/v3.1-1

In ChessGUI, your engine will run, but doesn't send its thinking line and eval to the GUI.

Graham Banks wrote: ↑Fri Feb 23, 2024 6:42 am In ChessGUI, your engine will run, but doesn't send its thinking line and eval to the GUI.

Yeah it's UCI interface is currently very minimal, I'll implement that real quick
Although I can't actually show the full PV line right now, as it's not being collected in Maxwell (I tried to implemented a Triangular PV-Table a couple versions ago, but it slowed down the engine dramatically so I cut it; I'll try again soon )

Are there any other UCI commands you'd like implemented?

Graham Banks wrote: ↑Fri Feb 23, 2024 6:42 am In ChessGUI, your engine will run, but doesn't send its thinking line and eval to the GUI.

Here's the new patch! https://github.com/eboatwright/Maxwell/ ... tag/v3.1-2

eboatwright wrote: ↑Fri Feb 23, 2024 4:24 pm

Graham Banks wrote: ↑Fri Feb 23, 2024 6:42 am In ChessGUI, your engine will run, but doesn't send its thinking line and eval to the GUI.

Here's the new patch! https://github.com/eboatwright/Maxwell/ ... tag/v3.1-2

Thanks. I'll try it shortly.

Maxwell now displays the depth and evaluation, but not its main thinking line.

Graham Banks wrote: ↑Fri Feb 23, 2024 7:02 pm Maxwell now displays the depth and evaluation, but not its main thinking line.

Maxwell currently doesn't collect the PV line, because the old PV-table implementation I had in development slowed down the engine alot, so I had to cut it and just print out the current best move instead.

I'll work on it

eboatwright wrote: ↑Fri Feb 23, 2024 7:14 pm

Graham Banks wrote: ↑Fri Feb 23, 2024 7:02 pm Maxwell now displays the depth and evaluation, but not its main thinking line.

Maxwell currently doesn't collect the PV line, because the old PV-table implementation I had in development slowed down the engine alot, so I had to cut it and just print out the current best move instead.

I'll work on it

I'll put Maxwell into Division 10.

Graham Banks wrote: ↑Fri Feb 23, 2024 8:01 pm I'll put Maxwell into Division 10.

Thank you so much!

Collecting the PV line was much much simpler than I thought, I went with the easier method of recursively walking through the transposition table
Here's the new patch: https://github.com/eboatwright/Maxwell/ ... tag/v3.1-3

eboatwright wrote: ↑Fri Feb 23, 2024 9:16 pm

Graham Banks wrote: ↑Fri Feb 23, 2024 8:01 pm I'll put Maxwell into Division 10.

Thank you so much!

Collecting the PV line was much much simpler than I thought, I went with the easier method of recursively walking through the transposition table
Here's the new patch: https://github.com/eboatwright/Maxwell/ ... tag/v3.1-3

Great stuff!

Graham Banks wrote: ↑Fri Feb 23, 2024 10:32 pm

eboatwright wrote: ↑Fri Feb 23, 2024 9:16 pm

Graham Banks wrote: ↑Fri Feb 23, 2024 8:01 pm I'll put Maxwell into Division 10.

Thank you so much!

Collecting the PV line was much much simpler than I thought, I went with the easier method of recursively walking through the transposition table
Here's the new patch: https://github.com/eboatwright/Maxwell/ ... tag/v3.1-3

Great stuff!

Thanks! I'm excited to see how this version fairs, it should be a considerable jump from v3.0.8-1!
I'd estimate about mid 2000s

Hello CCRL, just wanted to post a little update to say Maxwell is still in development!
I've been working on an NNUE implementation, and training an NNUE is proving to be even more difficult than I expected
Also if anybody's curious, the network architecture I'm working on now is: 768->128->1x8 which isn't big by any means, but it should be more than enough to be stronger than Maxwell's current HCE