TensorProb documentation

You’ve found the documentation for TensorProb, a probabilistic programming framework based on TensorFlow.

TensorProb is currently under construction! Expect things to break!

We are working on implementing the following features:

  • High flexibility in defining the statistical model
  • Models are defined in a self-contained with block
  • Seamless switching between frequentist and bayesian paradigms
  • Finding the maximum likelihood estimate or MAP estimate using a variety of optimizers
  • Flexible sampling using different MCMC backends
  • An extensive library of probability distributions
  • Analytic and numeric marginalization of probability distributions to support missing data and physical boundaries
  • Convolution of probability distributions
  • Functions for calculating confidence and credible intervals
  • Functions for hypothesis testing

Benefits of using TensorFlow as a backend include

  • Fast evaluation of the model using multiple CPU threads and/or GPUs
  • Defining new probability distributions using symbolic variables in Python
  • Possibility to write new optimized operators in C++ and load them dynamically

The API documentation