PYLFIRE

PYLFIRE performs likelihood-free inference by ratio estimation (LFIRE) for simulator-based models to estimate model parameters when likelihood calculations are infeasible.

Key Features:

• Penalized logistic regression: Implements penalized logistic regression to perform ratio estimation for likelihood-free inference.
• Likelihood-free inference by ratio estimation (LFIRE): Supports LFIRE methodology suitable for simulator-based models with intractable likelihoods.
• Statistical classifiers: Leverages statistical classifiers to convert inference problems into classification tasks for ratio estimation.
• Parameter outputs: Produces point estimates and posterior distributions for model parameters from simulated data.
• Integration with ELFI: Integrates with ELFI (Engine for Likelihood-Free Inference) for use within likelihood-free inference workflows.
• Implementation and requirements: Implemented in Python and requires Python 3.6 or higher and a Fortran compiler.

Scientific Applications:

• Population genetics: Inference of parameters in population genetics models where likelihoods are intractable.
• Astronomy: Parameter estimation in simulator-based astronomical models with complex data-generating processes.
• Economics: Application to economic simulator models for parameter inference without explicit likelihoods.
• Complex simulator-based models: General use for simulator-based models across disciplines when traditional likelihood calculations are infeasible.

Methodology:

Performs likelihood-free inference by training penalized logistic regression classifiers as statistical classifiers for ratio estimation (LFIRE) on simulated data to derive point estimates or posterior distributions, with integration via ELFI.