Benchmarks#
Runtime and statistical performance benchmarks for Latents methods.
These benchmarks serve two purposes:
User guidance — Help users understand computational scaling and data requirements for each method.
Correctness verification — Transparent proof that implementations recover ground truth parameters across a range of conditions.
Benchmarks are organized by method. Results are generated from pre-computed data to keep documentation builds fast.
Note
The scripts in this gallery require data files and dependencies not included in standalone downloads. To run benchmarks or regenerate results, see the Benchmarks page in the developer guide for setup instructions.
Group Factor Analysis (GFA)#
Benchmarks for Group Factor Analysis covering runtime scaling and parameter recovery.
Runtime benchmarks validate theoretical complexity:
Linear scaling in samples (N) and observed dimensions (D)
Cubic scaling in latent dimensions (K), though optimized LAPACK routines yield quadratic scaling at typical sizes
Recovery benchmarks verify parameter estimation across:
Sample sizes and noise levels
Observed and latent dimensionalities
Number of groups
Dimensionality recovery benchmarks evaluate post-hoc latent dimension selection across sample sizes and noise levels.
