Skip to content

endodod/sap-agent

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

sap-agent

A reinforcement learning agent that plays Super Auto Pets (Pack 1). The agent is trained with Maskable PPO (stable-baselines3 / sb3-contrib) inside a custom Gymnasium environment that wraps sap-sim, a full game simulation. After training, the agent plays 10,000 evaluation games and exports pick-frequency, win-rate, and synergy statistics consumed by sap-dashboard.

Related repos

  • sap-sim — the game simulation (Pack 1, v1.0)
  • sap-dashboard — Next.js dashboard that visualises the exported stats

Setup

# 1. Install the sim as a local dep
pip install -e ../sap-sim

# 2. Install agent deps
pip install -r requirements.txt

# 3. Generate the enemy pool (run once)
python scripts/gen_enemy_pool.py

# 4. Train
python agent/train.py

# 5. Evaluate and export stats
python agent/evaluate.py

TensorBoard logs are written to runs/ and model checkpoints to checkpoints/.

Exported data files

File Contents
data/enemy_pool.pkl Pre-generated pool of 500 opponent teams (one per round bucket). Required for training.
data/eval_results.json Per-game records from 10,000 evaluation episodes: outcome, rounds survived, lives remaining, every buy action, final team composition.
data/stats.json Aggregate statistics: overall win rate, mean rounds survived, per-pet pick frequency and win rate, top-20 synergy pairs, and win rate broken down by round reached.

About

Reinforcement learning agent that plays Super Auto Pets

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages