zhanjunxiong/fishbait

An AI agent designed to play poker.

Fishbait 🐟

Fully Insightful Superhuman Hold'em Bot with Artificially Intelligent Thought. An AI agent designed to play poker. Inspired by the Pluribus poker bot developed by CMU and Facebook.

Development Setup

From the root directory:

1. Create the build folder: mkdir -p build
2. Create the hyperparameters file: cp src/ai/mccfr/hyperparameters.h.dev src/ai/mccfr/hyperparameters.h
3. Navigate to the build folder: cd build
4. Generate the build files: cmake <options> ..
   • Possible Options:
     • -DCMAKE_C_COMPILER=<c-compiler> and -DCMAKE_CXX_COMPILER=<cpp-compiler>
     • -DCMAKE_BUILD_TYPE=Debug to debug with gdb or lldb.
     • -DCMAKE_BUILD_TYPE=Release to run in release mode. Release mode will add optimizations that makes the code run faster.
     • -DTGT_SYSTEM=Graviton2 for AWS Graviton2 optimizations.
     • -DCORES=<core_count> to instruct parallel algorithms to use <core_count> cores.
5. Build the code: cmake --build .
   • cmake --build . -- -j CORES to run a parallel build.
6. Download the card cluster files to build/out/ai/clustering
   • Flop
   • Turn
   • River
7. Generate the dev strategy: ./bin/blueprint.out
8. Convert the dev strategy to the hdf5 format: ./bin/avg_to_hd5.out out/ai/mccfr/dev/average_final.cereal out/ai/mccfr/dev/blueprint_dev.hdf
9. Create the api config file: cp ../src/api/src/settings.py.example ../src/api/src/settings.py
10. Start the API: ./bin/api.sh
11. Start the interface: ./bin/interface.sh

All executables will now be located in build/bin. So, for example, to execute poker_demo.out, simply run ./bin/poker_demo.out from the build directory.

Deployment

1. Build the project as detailed above.
2. Copy fishbait.nginx.example to /etc/nginx/sites-available/fishbait.nginx and replace the placeholders
   • Set the host property to the absolute path of build/src/interface/build
   • Set the server_name property to be whatever url the interface is to be deployed on.
3. sudo ln -s /etc/nginx/sites-available/fishbait.nginx /etc/nginx/sites-enabled/fishbait.nginx
4. sudo systemctl reload nginx
5. Configure src/api/src/settings.py
6. Copy src/api/fishbait.api.service.example to /etc/systemd/system/fishbait.api.service and replace the placeholders
   • Set the WorkingDirectory to the absolute path of src/api/src
   • Set the placeholder in ExecStart to be the absolute path of build/src/api/venv/bin/gunicorn
7. Copy src/interface/fishbait.depot.service.example to /etc/systemd/system/fishbait.depot.service and replace the placeholders
   • Set the WorkingDirectory to the absolute path of src/api/src
   • Set the placeholder in ExecStart to be the absolute path of build/src/api/venv/bin/python
8. sudo systemctl daemon-reload
9. sudo systemctl start fishbait.api
10. Setup https/ssl with the appropriate certbot instructions

Testing

Unit tests can be run with ./bin/tests.out from the build directory. If the project is built in debug mode, a code coverage report can be generated with make coverage from the build directory. gcov and lcov need to be installed for code coverage.

Requirements

• C++17
• CMake >= 3.20
• Gcov (for code coverage)
• LCOV (for code coverage)
• Python 3.10
• cpplint (runs during the pre-commit hook)
• pylint (runs during the pre-commit hook)
• nginx (to deploy the interface)

Other Notes

• Do not put commas in the Catch2 tester descriptions
• This project follows the Google C++ Style guide with the following exceptions:
  • Exceptions are allowed
  • Constructors are allowed to do work and fail
  • Integer types other than int are allowed
  • Type names with acronyms (like OCHS_N) may have underscores.
• This project follows the Google Python Style guide with the following exceptions:
  • 2 spaces are used for indentation
• Needs to be run on a processor where 32 bit integer reads and writes are atomic.