/gradient-descent-visualizer

A fast, interactive tool to visualize how different gradient descent algorithms (like vanilla gradient Descent, Momentum, RMSprop, Adam, etc.) navigate complex loss surfaces in real time.

Primary LanguageJavaScriptMIT LicenseMIT

🎯 Optimizer Visualizer

A dynamic 3D tool to explore how optimization algorithms traverse a loss surface in real time.

🌐 Live Demo

👉 Try it here

🎥 Demo Previews

3D Loss Surface Different optimizers Custom surface
Loss Surface optimizers Custom surface

✨ Features

  • 🧠 Explore optimizers: SGD, Momentum, Adagrad, RMSProp, Adam
  • 📈 Real-time 3D visualizations using Plotly
  • ✍️ Input custom loss surfaces using different functions
  • 🎚️ Adjust hyperparameters: learning rate, momentum, decay, beta values
  • ⏯️ Play, pause, reset animations and reinitialize points

🚀 How to Use

  1. Enter a loss function or select a default surface, e.g.:
    pow(w*w + b - 6, 2) + pow(w + b*b - 5, 2)
  2. Choose an optimizer and adjust its parameters
  3. Click or tap on the plot to place a starting marker
  4. Press Start to begin optimization
  5. Pause or reset to test different configurations

🙌 Acknowledgements

This project was built to help learners visually grasp the behavior of optimization algorithms used in deep learning and machine learning.