Tutorials

Interactive Jupyter notebook tutorials demonstrating various features of tgmm, numbered in the suggested reading order.

1. Foundations

• 1. Gaussian Mixture Models (GMM) - Basic introduction to fitting GMMs with tgmm, including a comparison of mean/weight/covariance initialization strategies
• 2. EM Algorithm Walkthrough - Step-by-step visualization of how the EM algorithm updates responsibilities and parameters each iteration
• 3. Visualization Techniques - Comprehensive guide to plot_gmm, the toolkit every later tutorial uses

2. Fitting Variants

• 4. Classification EM (CEM) - Comparing the hard-assignment CEM algorithm against standard EM
• 5. Supervised Fitting - Comparing EM, CEM, and label-supervised fitting on the same synthetic data

3. Evaluation & Utilities

• 6. Clustering Metrics - Evaluating GMM performance with various metrics
• 7. Sampling from GMMs - Generating synthetic data from fitted models
• 8. PCA Plotting - Visualizing high-dimensional GMM results with PCA

4. Bayesian Priors

• 9. Prior Distributions - Using priors for regularization and MAP estimation
• 10. NIW Priors Comparison - Detailed comparison of Normal-Inverse-Wishart priors

5. Advanced / Nonparametric Methods

• 11. Dirichlet Process Mixtures: Stick-Breaking and Gibbs Sampling - The math (stick-breaking, Chinese Restaurant Process), a step-by-step sampler walkthrough, then validating GaussianMixture(n_components=None, ...) against classical EM at scale
• 12. HDBSCAN Clustering - Density-based clustering with a PyTorch HDBSCAN implementation, including using it to remove noise before fitting a GMM

Running Tutorials

Option 1: Google Colab

Click the "Open in Colab" button at the top of each notebook.

Option 2: Local Jupyter

Install Jupyter:

pip install jupyter

Launch Jupyter:

cd docs/notebooks
jupyter notebook

Option 3: JupyterLab

pip install jupyterlab
jupyter lab

Requirements

All tutorials require:

pip install torch numpy matplotlib scipy
pip install tgmm  # or install from source

Some tutorials may have additional dependencies listed at the top of the notebook.