PYLFIRE: Python implementation of likelihood-free inference by ratio estimation.

Area of Science:

• Statistics
• Computational Science
• Machine Learning

Background:

• Simulator-based models are prevalent in diverse scientific domains, including population genetics, astronomy, and economics.
• Traditional statistical inference methods often struggle with models lacking an explicit likelihood function.
• Likelihood-free inference (LFI) has emerged as a key statistical methodology to address this challenge.

Purpose of the Study:

• To introduce PYLFIRE, an open-source Python implementation of the Likelihood-Free Inference by Ratio Estimation (LFIRE) method.
• To provide a user-friendly tool for performing parameter estimation in simulator-based models.
• To integrate PYLFIRE into the existing ELFI (Likelihood-Free Inference in Python) software ecosystem.

Main Methods:

• PYLFIRE utilizes the LFIRE method, which employs penalized logistic regression for parameter inference.
• The method leverages statistical classifiers to estimate model parameters or posterior distributions.
• Implementation is in Python, ensuring accessibility and ease of use for researchers.

Main Results:

• PYLFIRE offers an efficient and accessible open-source solution for likelihood-free inference.
• The tool facilitates the estimation of parameters for complex simulator-based models.
• Integration with ELFI enhances its utility for both users and developers in the LFI community.

Conclusions:

• PYLFIRE represents a valuable contribution to the field of likelihood-free inference.
• The open-source nature and integration with ELFI promote wider adoption and collaborative development.
• This tool empowers researchers to apply advanced statistical inference techniques to simulator-based models.