Analysis of Super Mario Bros gameplay data to identify scenes with interesting learning patterns across human players, imitation learning models, and PPO agents.
This project uses airoh for task automation and reproducible workflows.
# Install dependencies (includes airoh and invoke)
pip install -r requirements.txt
# Or use airoh task for setup
invoke setup
# (Optional) Install Jupyter kernel
pip install ipykernelTo see all available tasks:
invoke --listSee .invoke-cheatsheet.md for a quick reference of all tasks.
Option 1: Web Dashboard (Recommended - Fast & Modern)
# Using airoh task
invoke dashboard
# Or directly
python run_dashboard.py
# Opens at http://localhost:8050Option 2: Export to HTML (For Sharing)
# Export all scenes using airoh task
invoke export-html
# Export specific level or scene
invoke export-html --level=w1l1
invoke export-html --level=w1l1 --scene=0
# Or use the Python script directly
python export_dashboard.py --output-dir html_export
# Then open html_export/index.html in your browser
# Share the entire html_export/ folder with collaboratorsOption 3: Jupyter Notebook (Legacy)
jupyter notebook notebook/figure_exploration.ipynbGenerate metrics from raw data:
# Using airoh task (recommended)
invoke process-data
# Or directly
python code/make_df_metrics.pySee all available airoh tasks:
invoke --listKey tasks:
invoke setup- Install Python dependenciesinvoke process-data- Generate metrics from raw datainvoke dashboard- Launch interactive web dashboardinvoke export-html- Export scenes to standalone HTMLinvoke stats- Display dataset statisticsinvoke list-scenes- List all available scenesinvoke clean- Remove generated outputs
code/- Core processing and visualization modulesnotebook/- Jupyter notebooks for analysissourcedata/- Data files (parquet format)run_dashboard.py- Launch web dashboardexport_dashboard.py- Export to HTML files
See CLAUDE.md for detailed architecture documentation.