BayesLDM: A Domain-specific Modeling Language for Probabilistic Modeling of Longitudinal Data

Area of Science:

• Computational statistics
• Machine learning
• Biostatistics

Background:

• Longitudinal data analysis is crucial for understanding dynamic processes.
• Modeling complex multivariate time series presents significant computational challenges.
• Existing methods often require extensive programming expertise for efficient inference.

Purpose of the Study:

• To introduce BayesLDM, a library for Bayesian longitudinal data modeling.
• To provide a high-level modeling language and compiler for efficient probabilistic inference.
• To accelerate iterative modeling workflows for complex time series data.

Main Methods:

• Development of BayesLDM, a library featuring a high-level modeling language.
• Implementation of a compiler for generating optimized probabilistic program code.
• Focus on declarative specification of dynamic Bayesian Networks (DBNs).
• Integration of model specification with data inspection for inference code generation.

Main Results:

• BayesLDM enables efficient, declarative specification of DBNs.
• The compiler optimizes code for Bayesian inference and handles missing data.
• Demonstrated acceleration of iterative modeling workflows.
• Successful application to heterogeneous, partially observed mobile health data.

Conclusions:

• BayesLDM significantly simplifies and accelerates Bayesian longitudinal data modeling.
• The library abstracts the complexities of generating efficient probabilistic inference code.
• BayesLDM is a valuable tool for researchers analyzing complex time series data, particularly in mobile health.