Skip to content

Latest commit

 

History

80 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

PylaAI N — Personal Fork

Personal fork of PylaAI — a Brawl Stars bot. Extended with behavior trees, reinforcement learning, overlay, and a bunch of other stuff.

Setup

  1. Install Python 3.11 (tested with 3.11.9)
  2. pip install -r requirements.txt
  3. If you have an Nvidia GPU, install the GPU version of PyTorch for better performance: https://pytorch.org/get-started/locally/ (you'll need CUDA + cuDNN)
  4. python main.py

This is the localhost version — API features like login, online stats, auto-updating icons/models are disabled. You can change the api base url in utils.py if you want to hook it up to something.

Tests

python -m pytest tests/ -v

RL Monitor

Use rl_monitor.py to inspect PPO progress and verification metrics:

  • python rl_monitor.py (auto mode)
  • python rl_monitor.py --mode lenient
  • python rl_monitor.py --mode strict
  • python rl_monitor.py --mode auto --json
  • python rl_monitor.py --mode strict --exit-code
  • python rl_monitor.py --mode strict --exit-code --fail-on-warn
  • python rl_monitor.py --mode auto --exit-code --fail-on-insufficient-data
  • powershell -ExecutionPolicy Bypass -File .\scripts\run_survival_v2_smoke.ps1 -Mode auto
  • powershell -ExecutionPolicy Bypass -File .\scripts\run_survival_v2_smoke.ps1 -Mode strict -FailOnWarn
  • powershell -ExecutionPolicy Bypass -File .\scripts\run_survival_v2_smoke.ps1 -Mode auto -FailOnInsufficientData

Mode behavior:

  • auto: uses lenient below 3000 episodes, then strict
  • lenient: easier thresholds for early training
  • strict: tighter thresholds for stable long-run evaluation

Verification output can show INSUFFICIENT_DATA while episode/sample windows are still too small. Use --json for machine-readable output in scripts/dashboards. Use --exit-code to return process code 1 only when verification overall is FAIL (0 otherwise). Use --fail-on-warn together with --exit-code to treat WARN as failing too. Use --fail-on-insufficient-data together with --exit-code to treat INSUFFICIENT_DATA as failing too. run_survival_v2_smoke.ps1 combines text + JSON monitor checks and returns the monitor exit code.

License

CC BY-NC 4.0 — don't sell it.

Credits

This project is built on top of PylaAI. Thanks to everyone who made the foundation for this:

And all the other contributors who helped along the way.

Upstream repos:

About

No description, website, or topics provided.

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages