A playful experiment about human randomness, rhythm vs chaos — powered by tiny neural networks.
“You think you’re random. You’re not.”
Bonzo vs Jack the Dripper is a short, interactive Red / Black (0–1) game that explores how predictable human behavior really is.
There is:
- ❌ no strategy
- ❌ no winning goal
- ✅ just you, trying to be random
Behind the scenes, tiny neural networks learn your patterns — not cards.
This project is about human bias, not gambling.
Why the names?
-
🥁 Bonzo = John Bonham, Led Zeppelin's legendary drummer → rhythm, repetition, steady patterns
-
🎨 Jack the Dripper = Jackson Pollock, famous for drip painting → chaos, irregularity, anti-patterns
The AI measures things like:
- alternation vs repetition
- streak lengths
- short motifs (
001,010,111…) - how quickly you drift into rhythm
Before going to the browser, this project started as a Python + Tkinter PoC for:
- validate the UX
- test online learning stability
- tune window sizes & training strategy
- debug “predict once, train once” logic
Under the hood two tiny models run in parallel:
| Model | What it’s good at |
|---|---|
| GRU | Learning short & mid-term habits |
| Conv1D | Detecting repeating micro-patterns |
Their predictions are combined into an ensemble, with a live slider to mix their influence.
The models:
- predict once per click
- train online, continuously
- adapt quickly with as little as 30–70 inputs
- Append-only input (0 / 1, paste allowed)
- Conv1D + GRU models
- Ensemble mixing
- Real-time accuracy, SMA & cumulative plots
- Explicit TensorFlow memory management
pip install tensorflow numpy matplotlib pillow
python PoC-RedBlack-GRU-Conv1D.py📄 See 'PoC-RedBlack-GRU-Conv1D.py' for full implementation.
A fully browser-based implementation using TensorFlow.js for maximum accessibility.
Live Demo: 👉 Try it here!
- 🌐 Zero Installation: Runs directly in any modern web browser
- 📱 Mobile Friendly: Responsive design with touch support
- 🎨 Polished UI: Smooth animations and professional styling
- 📊 Real-time Charts: Chart.js integration for live performance graphs
- 🎯 AI Models: Conv1D and GRU networks implemented in TensorFlow.js
- ⚖️ Ensemble Mixing: Interactive slider to blend predictions
- 🖼️ Visual Theme: Card-based interface with custom graphics
- TensorFlow.js 4.15.0: Neural network inference in the browser
- Chart.js 4.4.1: Real-time performance visualization
📄 See `index.html' for full JS implementation.
Pro tip: True randomness is harder than you think! Most people unconsciously alternate or create repeating patterns.
Good luck! 🍀