New engine: Stash

[JohnW]

What's the estimated rating now? Seems a lot stronger than 1450

[CMCanavessi]

Around 3000

[mvanthoor]

Good luck with version 28.

I like Stash's coding style, with modularized parts and small files. It looks a lot like how Rustic is structured. (And thus how I'd do it if I had written the engine in C or C++.) Is there running changelog available, where you keep all the changes from one version to the next, and how much strength they brought?

[mhouppin]

Good luck with version 28.

I like Stash's coding style, with modularized parts and small files. It looks a lot like how Rustic is structured. (And thus how I'd do it if I had written the engine in C or C++.) Is there running changelog available, where you keep all the changes from one version to the next, and how much strength they brought?

Hi Marcel, I'm working on it. I'm currently retrieving all the results from my SPRT runs on OpenBench, and adding a Changelog to the repository with all the information available about how much each patch brings Elo. I don't have any data prior to version 22 (since I deleted the local OpenBench instance I used), so I planned to redo SPRT runs on the old patches (this shouldn't take too long since I rarely committed anything under +10 Elo during this period).

I just added the first Changelog version to the repository for version 28, older versions coming soon to the file. Good morning !

[mhouppin]

Hi there, version 29 is out ! It has already been reported in the "Engine Releases 2021" thread, but I'll list the improvements made:

- Added staged move generation to move picking;
- Changed the place_top_move() function to only swap the top move and the best move instead of shifting the whole list;
- Placed SEE-negative captures after quiets in search;
- Added SEE pruning to search;
- Changed Late Move Pruning code to stop pruning captures.

Note: there is a known bug with underpromotions in search for this version in particular, it won't just generate them.
This bug doesn't affect Elo much, and doesn't break the engine (it still accepts underpromotions when setting up the position), but if some people intend to use Stash as an analysis engine, development version doesn't have the issue anymore, so you might want to pick this one.

This release has scored +58 Elo against version 28, and I expect the "real" gains to be about 40 Elo.
The binaries are available here: https://gitlab.com/mhouppin/stash-bot/-/releases/v29.0

Good evening to everyone !

[Modern Times]

Hi there, version 29 is out ! It has already been reported in the "Engine Releases 2021" thread, but I'll list the improvements made:

This release has scored +58 Elo against version 28, and I expect the "real" gains to be about 40 Elo.
The binaries are available here: https://gitlab.com/mhouppin/stash-bot/-/releases/v29.0

It was tested here:

http://ccrl.chessdom.com/ccrl/404FRC/cg ... _length=30

[mhouppin]

Hi there, version 29 is out ! It has already been reported in the "Engine Releases 2021" thread, but I'll list the improvements made:

This release has scored +58 Elo against version 28, and I expect the "real" gains to be about 40 Elo.
The binaries are available here: https://gitlab.com/mhouppin/stash-bot/-/releases/v29.0

It was tested here:

http://ccrl.chessdom.com/ccrl/404FRC/cg ... _length=30

Wow, this suggests even larger gains that what I expected. Great news ^^

[mhouppin]

Hi there, version 30 is out ! Here's the list of the changes made: https://gitlab.com/mhouppin/stash-bot/- ... 2021-04-30

Since most of the improvements are from search, the +51 Elo I got in regression tests might well be smaller against other engines. I expect to fix that for v31.

Note: the Gitlab repo will get some cleanups soon, because I don't want it to contain thousands of branches from OpenBench tests. From now, I'll run the tests on Github only, then make the merges on the Github master branch and push them back on the Gitlab repo, effectively keeping only one branch there. If you still need all the branches for some reason, they will be available here: https://github.com/mhouppin/stash-bot

Good afternoon to everyone ! ^^

[Modern Times]

Hi there, version 30 is out ! Here's the list of the changes made: https://gitlab.com/mhouppin/stash-bot/- ... 2021-04-30

Since most of the improvements are from search, the +51 Elo I got in regression tests might well be smaller against other engines. I expect to fix that for v31.

I got +22 Elo at chess960 but not a huge number of games:

http://ccrl.chessdom.com/ccrl/404FRC/cg ... _length=30

[Chessqueen]

Hi there, version 30 is out ! Here's the list of the changes made: https://gitlab.com/mhouppin/stash-bot/- ... 2021-04-30

Since most of the improvements are from search, the +51 Elo I got in regression tests might well be smaller against other engines. I expect to fix that for v31.

Note: the Gitlab repo will get some cleanups soon, because I don't want it to contain thousands of branches from OpenBench tests. From now, I'll run the tests on Github only, then make the merges on the Github master branch and push them back on the Gitlab repo, effectively keeping only one branch there. If you still need all the branches for some reason, they will be available here: https://github.com/mhouppin/stash-bot

Good afternoon to everyone ! ^^

Thanks for releasing such great engine, which is one of the faster progressing engine
https://gitlab.com/mhouppin/stash-bot/-/releases/v30.0