bayesloop

Efficiently infers parameters that change over time, as demonstrated in the coal mining disaster example where the disaster rate evolves gradually, enabling analysis of non-stationary systems.

Uses Bayesian model evidence to compare hypotheses about parameter dynamics, providing a statistical basis for model comparison without relying on subjective criteria, as highlighted in the documentation.

Employs a forward-backward algorithm to process data step-by-step, avoiding the curse of dimensionality for high-dimensional parameter spaces, making it scalable for complex dynamic systems.

Allows building custom models based on SymPy and SciPy, enabling researchers to tailor probabilistic models to specific scientific questions, such as in cancer research or financial risk assessment.