PyMC: a modern, and comprehensive probabilistic programming framework in Python

Area of Science:

• Statistics
• Computer Science
• Machine Learning

Background:

• Probabilistic programming enables complex statistical model construction.
• Bayesian methods are crucial for uncertainty quantification.
• Efficient model fitting requires optimized computational backends.

Purpose of the Study:

• Introduce PyMC, a versatile Python library for Bayesian modeling.
• Showcase PyMC's capabilities in fitting diverse statistical models.
• Highlight PyMC's contribution to the open-source probabilistic programming community.

Main Methods:

• Utilizes PyTensor for symbolic computation and compilation.
• Supports multiple computational backends (C, JAX, Numba).
• Leverages various hardware architectures (CPU, GPU, TPU).

Main Results:

• Demonstrates ease of use and versatility across common statistical models.
• Facilitates fitting of generalized linear models, time series, ODEs, and Gaussian processes.
• PyMC enables efficient Bayesian inference on diverse computational hardware.

Conclusions:

• PyMC provides an intuitive and powerful framework for Bayesian analysis.
• Its flexible architecture supports a wide range of statistical modeling tasks.
• PyMC plays a significant role in advancing open-source probabilistic programming tools.