BSTPP: a python package for Bayesian spatiotemporal point processes.

Area of Science:

• Computational statistics
• Geospatial analysis
• Event data modeling

Background:

• Spatiotemporal point process models are effective for analyzing event data but are often difficult to implement.
• A lack of accessible Python packages hinders their application, particularly for Bayesian inference.
• Existing methods require significant programming expertise, limiting broader adoption.

Purpose of the Study:

• To present BSTPP, a novel Python package designed for Bayesian inference on spatiotemporal point processes.
• To provide an accessible and extendable framework for implementing various point process models.
• To facilitate the application of advanced statistical models to real-world event data.

Main Methods:

• BSTPP implements three core models: space-time separable Log Gaussian Cox, Hawkes, and Cox Hawkes processes.
• The package features an extendable Trigger module for custom parameterizations in Hawkes models.
• Posterior inference for Gaussian processes in Cox models is accelerated using a pre-trained Variational Auto Encoder (VAE).

Main Results:

• The BSTPP package offers a user-friendly interface for complex spatiotemporal point process modeling.
• The integrated Variational Auto Encoder (VAE) significantly speeds up Gaussian process inference.
• Simulation studies validated the model's performance, and its utility was demonstrated on Chicago shooting data.

Conclusions:

• BSTPP democratizes the use of sophisticated spatiotemporal point process models in Python.
• The package's flexibility and efficiency, particularly with the VAE, address key implementation challenges.
• BSTPP is a valuable tool for researchers analyzing event data across various domains.