DSchack 2.0: now with cool new move generator!
DSchack is the chess engine held together by duct tape, hopes and prayers,
and many long nights at the so called "Hackkvällar" at my local university.
DSchack is released under the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or (at your
option) any later version.
Questions
=========
Q: What is this?
A: This is a chess engine that I am developing as a hobby project. It
implements the UCI protocol so you can play against it in any GUI
which supports that.
Q: What's the engine's playing strength?
A: Probably very weak compared to Stockfish, but probably stronger
than most human players.
Q: Hardware requirements?
A: The engine probably works best on 64-bit x86 family processors
with fast support for PEXT and PDEP instructions, which the move
generator relies heavily upon, but ideally the engine should
compile and function (albeit slowly) on most architectures.
Otherwise please open an issue, and if you want to, feel free to
dive into the code yourself to try to fix it.
Q: Why is the code indented so weirdly?
A: Set your tab width to 8.
(GNU Emacs' default formatting is weird and I don't care enough
to bother figuring out how to configure it to something else.)
Q: What does DSchack mean?
A: "Schack" is Swedish for "Chess" and the letter 'D' in there can
mean a lot of different things depending on my mood a particular
day. If I'm feeling egoistic, it can be the initial letter of my
first name. Occasionally it stands for "dumb".
Roadmap
=======
This is a list of concepts and ideas that I want to implement or
experiment with in the future. This is only an approximate list of
my plans for this engine, as those plans change very frequently.
- Killer heuristic.
- Transposition table.
Maybe replace the PV-table-backed functionality entirely and
probe the PV directly from the TT instead.
- Neural networks for evaluation.
I want to experiment with my own ideas for how to train a neural
network to evaluate chess positions, taking some inspiration from
ideas found in algorithms such as Deep Q-learning (DQN) and
Proximal Policy Optimization (PPO). This will probably differ
considerably from how NNUE-like evaluators are typically trained,
maybe for the better or maybe for the worse.