Jimbo - Chess Engine

Minor code cleanups

1. Switched opening book to Cerebellum, which is WAY better than the previous one
2. Fixed an issue in the time taken to play moves for the lichess bot.

Will ship soon!

1. Internal logic improvement - use Unmake move, instead of storing a copy of each board.
2. Use compiler intrinsic blsr, instead of the previous, inefficient alternative.

Engine now stops search after it finds a forced checkmate, cutting out the delay in those situations.
Currently getting my bot rated on lichess, so it’s playing matches against as many other bots as possible. Will enable rated matches soon.

Made move searching a separate thread, so now you can pass the UCI “stop” command to terminate the search at that point and return the best move!

1. Cleaned up and reduced the file count
2. Improvements in SPEEEEEED(therefore performance)!
3. Outputs “Illegal command” when required, rather than ignoring.

Refactored code to use a standard build (with .h header files) instead of my old “unity” build method.

Completely refactored and revamped the Lichess bot code! Now with -

1. Proper logs with python logging
2. Handle multiple games without freezing
3. Accept all time controls
4. Use streaming instead of polling, comes with massive latency improvements.
5. Use the standard chess.SimpleEngine to handle engines, so multiple engines work with it.

Updated the Lichess Bot script for better usability.

Added builds for windows and macos, go check out the latest release @ https://github.com/abhijato-c/Chess-Engine/releases !

Added magic bitboards implementation, so now it can compile on CPU’s without BMI2 architecture!
Now working on compiling all binaries and releasing them in a manner similar to how stockfish releases.
Another match won against stockfish level 4, this time with the new magic bitboard implementation! (I should seriously start playing against higher levels now)

Added a few lines that will allow this to compile (mostly)everywhere. All it needs now is support for magic bitboards, as the pext instruction(current implementation) is available only on certain cpu’s.
Will work on that tomorrow.
Attached screenshot below of Jimbo crushing Stockfish lvl 4!(after many losses, that is)

Added a file MakeOpeningBook.py that parses all PGN files in the ‘pgns’ directory, and makes an opening book from them.
I have downloaded around 250 PGN files for cow, containing a whopping 500K games from pgn mentor! This makes my opening book go from around 10mb to ~30mb, and I hope this version is more comprehensive.
Attached yet another screenshot of matches against stockfish for testing.

Minor speed & efficiency improvements. I’m burned out working on this engine, so I will now spend some time improving the opening book, and working on other projects.

Officially the longest I’ve spent hunting down a bug. A verry tiny and nearly insignificat bug in the evaluation code took me 6 entire hours to resolve. The original plan was for me to add castling to the bot(yes, the bot can’t castle yet), but looks like that’s postponed to tomorrow. For now, at least, the bot plays ever so slightly better.
Attached below is the one of dozens of games I made the bot play trying to debug…

Edited the bot to accept only unrated correspondence challenges. Code is now running on google collab, so it should be online more often.
Working on making it accept more time controls.
Challenge at https://lichess.org/@/M8000_bot with correspondence time control and casual, not rated.

Realized I hadn’t touched the Lichess API bot code for a while, so I improved it a bit, and made it more readable!
Also ran some additional tests, and realized that tho it wins pretty consistently against stockfish level 4, it loses REALLY bad against lvl 5 ;c
I know exactly why this is happening(Static evaluation needs to be done using positional advantages, not just absolute material advantages), but fixing it makes my engine perform much WORSE as it reduces performance.

Added some UCI support to make it standardized!
The support is still minimal, and it only handles a fraction of commands, but I will keep adding more commands in the future.

Found bug that was causing engine to play bad. The move generation code for knights had a typo in it, causing it to not generate all possible moves for the knight. This caused the engine to “forget” that the knight can move in a certain way, therefore missing crucial moves.
Now can beat stockfish level 4 consistently when given 3 seconds per move.
Pasted image of a game that it had previously blundered away, but now won(when given the position where it started losing pieces)

Cleaned up some buggy code and got the engine running on Lichess, it performs decently well. Beat Stockfish level 4 at 3 seconds per move, but otherwise, performance is underwhelming. Looking into ways to make the engine stronger. Possible inefficiencies in iterative deepening code.

This is an old, buggy project that I recently started fixing. It is a really basic chess engine that can play on lichess bot API via python.