Interactive statistics, machine learning, and signals explainers built with vanilla JavaScript, D3, Three.js, and MathJax. Each page pairs a live sandbox with step-by-step notes, pseudocode, and worked examples.
Distribution Visualizer (pages/distributions.html)
Add and compare probability distributions side by side.
Adjust parameters in real time and see the PDF, CDF, and key statistics update instantly.
Unsupervised Supervised Learning (pages/generative_classification.html)
Builds intuition for generative classifiers by showing how class-conditional densities produce decision boundaries.
Linear Regression Regularization (pages/regularization.html)
Visualize how L1 and L2 regularization shape the loss surface and shrink coefficients.
Drag data points to see how the fit responds.
Gradient Descent (pages/gradient-descent.html)
Compare full-batch, mini-batch, and stochastic gradient descent, and SGD, momentum, AdaGrad, RMSProp, and Adam optimizers, on a shared loss surface.
Estimation (pages/estimation.html)
Interactive demonstration of bias, variance, MSE, and MLE.
Sample repeatedly from a true distribution and watch the estimator's sampling distribution form.
Bayesian Inference (pages/bayesian.html)
Step through prior, likelihood, and posterior for common conjugate families.
Includes MAP estimation and an introduction to approximate inference.
Principal Component Analysis (pages/pca.html)
3D scatter plot with interactive camera, PC arrows, and a linked operator view showing the covariance ellipsoid.
Controls for dataset shape, noise, and dimensionality.
Fourier Image Decomposition (pages/fourier.html)
Upload an image and watch it reconstructed frequency by frequency.
Explains periodicity and the checkerboard artifact in the frequency domain.
Manifold Learning (pages/manifold.html)
Step-by-step comparison of LLE and Isomap on a shared dataset.
Each step shows the algorithm state as a 3D visualization with intuition, formula, and pseudocode panels.
Attention (pages/attention.html)
Single-head scaled dot-product attention worked end to end, from Q/K/V projections through masking, softmax, and the weighted sum.
A clickable pipeline diagram where every step expands into the real numbers behind it, with interactive worked examples you can steer by picking which token to follow.
Glossary (pages/glossary.html)
Reference definitions for terms used across the site.
Settings (pages/settings.html)
Theme and display preferences, saved on-device.
The pages use ES modules and must be served over HTTP, not opened as file:// URLs.
npx serve .Then open http://localhost:3000/pages/<page>.html.
- D3 v7 for 2D charts and SVG visualizations
- Three.js for 3D scenes on the PCA page
- MathJax 3 for typeset mathematics
- No build step; all pages are static HTML with ES module scripts