v0.2.0

UMAP.jl v0.2.0

Planned Features

• Handle arbitrary input data and multiple "views" of the dataset
• Transform (first available in v0.1.6)
• Multithreaded KNN graph construction
• API overhaul: basic API and advanced API

Planned Improvements

• Improve efficiency of optimization via make_samples_per_epoch logic
• Support passing a KNNGraph as precomputed distances
• Improve documentation for new and existing features

Tentative Features

• User-provided initial embedding
• Partial missing data support
• Supervised UMAP
• Semi-supervised UMAP (target labels with partial missing data)
• Parallel SGD
• Sparse inputs (vector of sparse vectors)

Not Planned for v0.2.0

• Inverse transform
• Freezing selected points during optimization (extension to user-provided initial embedding)
• Support embedding to arbitrary manifolds