Represent observations and choose their geometry | Alignment and relational geometry | Gaussian parameters and distribution summaries | Read the available evidence

Transport cost and Wasserstein distance | A small two-dimensional example | Supply cross-distances explicitly | Numerical scale and further workflows | References

One observation is a probability distribution | A cost and a distance have different units | Entropy regularization changes the objective | Scalar barycenters and medians | Choose a conversion deliberately | Work directly with Gaussian parameters

Numerical audit repairs in 0.2.0.9002 | Additive toolbox interfaces in 0.2.0.9001 | Recheck the target and numerical outcome | ECDF summaries preserve weighted finite distributions | Histogram displays and estimates are separate | Historical regularized names now share one checked implementation | Image methods have explicit grids and projection error | Gaussian means, variances, and covariance domains

Compare targets on the same observations | State what is allowed to change | Initialization and support-size sensitivity | Read failures and stopping criteria | Fixed-grid summaries

Event-related covariance matrices | Gorilla skull shape

Validate observations and select a geometry | Inspect a numerical summary | Fit, predict, and reconstruct tangent components | Reuse a clustering model | Tune scalar-response regression using every held-out fold | Preserve the fitted state

Existing names and result fields | Numerical controls are honored | Old regression objects without geometry | Refit old tangent and clustering objects | Domain changes are explicit

Example : 1d signal with denoise1 | Example : image denoising with denoise2

1. Install and Load | 2. Intrinsic Dimension Estimation | 3. Dimension Reduction

1. Load | 2. Computing Distances | 3. One Application : Embedding Networks, Not Network Embedding