bjpmodel 0.2.1

BJP (Bayesian Joint Probability)

Bayesian Joint Probability (BJP)

Table of Contents

• Overview
  • Installation
  • Usage
  • Contributing
  • License
  • Contact

---

Overview

BJP is a high-performance Python package (with Cython/C++ backend) for fitting multivariate normal distributions using Gibbs sampling, with support for left and right censoring. Typically, BJP is used for modelling with transformed variables using the norm-transform package. Therefore, it is designed for statistical modeling and forecasting in scenarios where data may be treated as partially observed (e.g. rainfall).

---

Features

• Fast Gibbs sampling for multivariate normal distributions
• Handles left and right censoring (if required)
• Generates ensemble predictions with posterior predictive sampling
• Cython/C++ backend for performance

---

Installation

Recommended: Using a virtual environment and installing into. A python virtual environment is a self-contained directory that contains a Python installation for a particular version, along with additional packages. It allows you to isolate dependencies for different projects.

Use the package manager pip to install normtransform.

python -m venv venv # Create a virtual environment
source venv/bin/activate  # Activate the virtual enviroment, on Windows use `venv\Scripts\activate`
pip install bjpmodel #install bjpmodel

Note: See compatible versions

---

Example Usage

This demo.py can be used as example for BJP. Once a python environment has been build and activated you can run demo.py from the commandline with python ./bjpmodel/demo.py

python ./bjpmodel/demo.py
[-1.30846209 -1.1149359 ] [2.31631585 0.69101736]
iterations: 1240
iterations: 1363
        The number of variables is 2
        The length of the Markov chain is 5000
        Sampling using 1000 time periods
Warning: Predictor is NaN or missing
--- Left and right censoring ---
Mean Data: [ 0.48266767 -0.19992095] Std Data: [1.17708839 0.57420349]
Mean Predictions: [ 0.4721382  -0.19274477] Std Predictions: [1.16840932 0.57216459]

---

Contributing

Requests are welcome. Please contact the authors below.

---

License

MIT

---

Contact

• Andrew Schepen - [e-mail]
• David Robertson - [e-mail]