ABOUT JUNGLANSVIS
Watch machine learning come alive.
JunglansVis (https://junglans.in) bridges the gap between mathematical equations and physical intuition. Instead of reading formulas in a vacuum or looking at static chart outputs, you watch algorithms explore parameter spaces, partition decision boundaries, and converge — in real time, step by step, right inside your browser.
The name & metaphor — Junglans regia is the botanical Latin name for walnut: a seed that looks remarkably like a miniature human brain and grows layer by layer, one tree ring at a time. We believe human intuition about machine learning grows in the exact same organic way — by experimenting with levers and watching systems learn from scratch.
What You Can Do in JunglansVis
- Explore 22 core machine learning algorithms spanning Classification, Regression, Clustering, Dimensionality Reduction, and Reinforcement Learning.
- Choose from 10 procedural synthetic datasets (Two Moons, Concentric Rings, Anisotropic Blobs, Sine Waves) or upload custom CSV datasets with automatic feature detection.
- Tune hyperparameters live via responsive sliders and inspect instantaneous visual updates within 16 milliseconds.
- Watch training unfold frame-by-frame with step controls or let the engine auto-play at configurable iteration speeds.
- Execute head-to-head experiments in dual Comparison Mode (Model A vs Model B on identical data).
- Inspect live loss and accuracy convergence curves plotted on device-scaled 2D Canvas graphs.
- Receive real-time expert diagnostic fitting tips (Underfitting, Good Fit, Overfitting) with pedagogical advice.
- Export publication-ready, high-DPI PNG snapshots of decision surfaces with a single click.
- 100% pure client-side TypeScript execution — zero install, zero server compute latency, and completely private.