Custom Distributions API Reference¶
Custom distribution implementations for SciStanPy models.
This submodule provides enhanced and custom distribution implementations that extend the capabilities of standard PyTorch and SciPy distributions. These distributions address specific modeling needs and limitations in the standard libraries while maintaining compatibility with SciStanPy’s multi-backend architecture.
The distributions maintain the standard PyTorch and SciPy interfaces while providing additional functionality needed for advanced probabilistic modeling scenarios. They are designed to be drop-in replacements for standard distributions in most cases, with enhanced capabilities for specific use cases.
There are two submodules for custom distributions, one holding distributions implemented using PyTorch and another holding distributions implemented using SciPy: