Changelog#

All notable changes to this project will be documented here.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased #

Documentation hosted on Read the Docs with interactive examples gallery and benchmark results
Python 3.14 support
observation/ subpackage with full probabilistic hierarchy: hyperpriors, priors with sample() methods, posterior classes, and realization containers
state/ subpackage with probabilistic hierarchy: LatentsPriorStatic, LatentsPosteriorStatic with sample() and posterior_mean, and LatentsRealization
plotting/ subpackage: hinton_diagram() for matrix visualization, plot_dimensionalities(), plot_var_exp(), and pair-plot variants for observation model results
GFAFitConfig and GFASimConfig frozen dataclasses for immutable configuration
GFASimulationResult dataclass bundling simulation outputs with persistence functions (save_simulation(), load_simulation(), save_simulation_recipe(), load_simulation_recipe())
Callback system for model fitting: ProgressCallback (tqdm), LoggingCallback (structured logging), CheckpointCallback (periodic safetensors checkpoints)
GFAModel.recompute_latents(Y) and GFAModel.recompute_loadings(Y) to restore cleared parameters after loading saved models

Package reorganized: observation_model/ → observation/ and latents.data, state_model/ → state/, plotting.py → plotting/, internal utilities → _internal/
GFA fitting loop optimized to reduce temporary memory allocations via np.einsum and pre-computed intermediate arrays. 10–65% faster per-iteration runtime depending on problem dimensions.
GFA inference iteration order changed from d → X → C → α → φ to d → C → α → φ → X, enabling exact latent reconstruction from saved observation parameters (resume_fit() and compute_lower_bound() now work with save_x=False)
simulate() signature changed to simulate(config, hyperprior) returning GFASimulationResult
Methods with in_place=True now return self instead of None
random_seed now accepts int | Sequence[int] | None for structured seeding in parallel experiments
Minimum Python version raised from 3.9 to 3.10

Python 3.9 support
pandas dependency
GFAFitArgs class (replaced by GFAFitConfig), HyperPriorParams (replaced by ObsParamsHyperPrior), StateParamsStatic (absorbed into LatentsPosteriorStatic)
verbose parameter from fitting config (use ProgressCallback instead)
notebook optional extra; install Jupyter separately if needed