TKinter-based K-Nearest Neighbors (tk2NN)

tk2NN is an interactive visualization of the k-nearest neighbors (k-NN) algorithm built using Tkinter. It lets you tweak the number of labels, points, and the value of k, and watch how the classification map changes in real time. You can toggle between Euclidean and Manhattan distance metrics, and colors for each label are randomized every time the graph is redrawn - even with the same points. Don’t like the current color scheme? Just re-select the same distance type. Boom. Fresh new look.

Features

• Switch between Euclidean and Manhattan distances
• Rerun classification instantly with updated parameters
• Random color assignments for each label every time
• Clear all points with a right-click
• Visualizes decision boundaries pixel-by-pixel
• Lightweight Tkinter GUI — no heavy dependencies
• Fully customizable via config.yaml — canvas size, colors, default parameters

Getting Started

Requirements

• Python 3.8+ (recommended)
• colour
• Pillow

Install the required packages:

bash pip install -r requirements.txt

Run the App

bash python run.py

If you're using Linux and want to install Python 3.8 explicitly:

bash sudo add-apt-repository ppa:deadsnakes/ppa sudo apt-get update sudo apt-get install python3.8 python3.8 -m venv venv source venv/bin/activate pip install -r requirements.txt python run.py

Project Structure

tk2nn/ ├── app/ # Core modules │ ├── gui.py # Main Tkinter app │ ├── knn_canvas.py # k-NN algorithm and classification logic │ └── utils.py # Random point generation, color logic, multimode fallback │ ├── assets/ # Images and visual resources │ ├── bg.png │ ├── demo.png │ └── multidemo.png │ ├── config.yaml # Centralized configuration ├── run.py # Entry point to launch the app ├── requirements.txt ├── README.md └── .gitignore

Notes

• The color scheme is randomized every time you update or re-select a distance metric.
• This project was inspired by another GUI project TKayTint (same developer — different beast).
• It’s a simple educational tool that can be extended to explore classification boundaries, real datasets, or even point dragging and manual labeling.

Future Ideas

• [ ] Add click-to-label custom point placement
• [ ] Export the canvas as a PNG snapshot
• [ ] Add more distance metrics (Chebyshev? Cosine?)
• [ ] Toggle persistent vs. random color palettes
• [ ] Performance boost with NumPy or shaders (if you’re crazy)

Contributing

Got ideas? Bugs? Feature requests? Feel free to: - Fork the repo - Submit a pull request - Or just open an issue and yell into the void

License

This project is licensed under the MIT License. See the LICENSE file for more info.