Articles
Practical guides
Current examples for configuring and running the package.
- Using smallvis
Install smallvis, make an embedding, plot it, and inspect the result.
- Inputs and initialization
Choose the input representation, preprocessing, neighborhood size, and starting coordinates.
- Optimization and repeated runs
Configure methods and optimizers, use exaggeration, and compare repeated starts.
- Barnes-Hut and native acceleration
Run Barnes-Hut embeddings, choose neighbor affinities, and control native worker threads.
Theory and background
Cost functions, spectral methods, datasets, and the motivation for smallvis.
- Some Theory
- Spectral Methods
- Datasets
- About smallvis
The motivation for smallvis and notes from earlier stages of its development.
Inputs and initialization
Experiments with preprocessing, neighborhoods, and starting coordinates.
- Input Scaling
- PCA and Whitening
- t-SNE Initialization Options
- Averaging Multiple Results
- Perplexity
- Perplexity and Intrinsic Dimensionality: Theory
- Perplexity and Intrinsic Dimensionality: Application to t-SNE
- Perplexity and Intrinsic Dimensionality: Non-Global Perplexities
- Multiscale perplexities
- Input Normalization in SNE
Embedding methods
Derivations and experiments with the objectives available in smallvis.
Optimization
Comparisons of optimizers, exaggeration, and optimization settings.
Comparative experiments
PaCMAP, NCVis, and investigations using synthetic datasets.