Tutorials#

Each tutorial is a runnable Colab notebook. Click the badge to open it directly in Google Colab — no local setup required.

Decision Geometry#

decision_geometry.ipynb Open In Colab

Real-world datasets (Wine, Breast Cancer, Digits, Iris) with multiple model families. Compares PCA and sensitivity projection side by side.

Latent Space Explorer#

latent_space.ipynb Open In Colab

Document embeddings, protein family vectors, and market regime states — all in 64–128 dimensions. PCA, t-SNE, and UMAP projections compared.

Synthetic Decision Boundaries#

synthetic_boundaries.ipynb Open In Colab

Two moons, concentric circles, anisotropic blobs, and XOR — each embedded in 20-D noise. Demonstrates sensitivity projection recovering the informative axes from pure noise.

High-Dimensional Feature Spaces#

high_dimensional.ipynb Open In Colab

Olivetti faces (4 096-D), 20 Newsgroups TF-IDF (300-D), and MNIST digits (784-D). Shows sensitivity projection on real high-dimensional data.

Model Comparison#

model_comparison.ipynb Open In Colab

Six model families — Logistic Regression, Linear SVM, RBF SVM, Random Forest, Gradient Boosting, MLP — trained on the same dataset and visualised side by side in both sensitivity and PCA frames.

Geographically Weighted PCA — Synthetic#

gwpca.ipynb Open In Colab

GWPCA implemented from scratch on a spatially structured synthetic dataset. Bandwidth sensitivity analysis and comparison with global PCA.

Geographically Weighted PCA — NYC Airbnb#

gwpca_nyc_airbnb.ipynb Open In Colab

GWPCA on the real Kaggle NYC Airbnb dataset (48 895 listings). Local PC explained variance mapped across the five boroughs; loading interpretation reveals borough-specific market dynamics.