TensorFlow Probability

TensorFlow Probability is a library for probabilistic reasoning and statistical analysis in TensorFlow. As part of the TensorFlow ecosystem, TensorFlow Probability provides integration of probabilistic methods with deep networks, gradient-based inference via automatic differentiation, and scalability to large datasets and models via hardware acceleration (e.g., GPUs) and distributed computation.

Our probabilistic machine learning tools are structured as follows.

Layer 0: TensorFlow. Numerical operations. In particular, the LinearOperator class enables matrix-free implementations that can exploit special structure (diagonal, low-rank, etc.) for efficient computation. It is built and maintained by the TensorFlow Probability team and is now part of tf.linalg in core TF.

Layer 1: Statistical Building Blocks

• Distributions (tfp.distributions, tf.distributions): A large collection of probability distributions and related statistics with batch and broadcasting semantics. See the Distributions Tutorial.
• Bijectors (tfp.bijectors): Reversible and composable transformations of random variables. Bijectors provide a rich class of transformed distributions, from classical examples like the log-normal distribution to sophisticated deep learning models such as masked autoregressive flows.

Layer 2: Model Building

• Edward2 (tfp.edward2): A probabilistic programming language for specifying flexible probabilistic models as programs. See the Edward2 README.md.
• Probabilistic Layers (tfp.layers): Neural network layers with uncertainty over the functions they represent, extending TensorFlow Layers.
• Trainable Distributions (tfp.trainable_distributions): Probability distributions parameterized by a single Tensor, making it easy to build neural nets that output probability distributions.

Layer 3: Probabilistic Inference

• Markov chain Monte Carlo (tfp.mcmc): Algorithms for approximating integrals via sampling. Includes Hamiltonian Monte Carlo, random-walk Metropolis-Hastings, and the ability to build custom transition kernels.
• Variational Inference (tfp.vi): Algorithms for approximating integrals via optimization.
• Optimizers (tfp.optimizer): Stochastic optimization methods, extending TensorFlow Optimizers. Includes Stochastic Gradient Langevin Dynamics.
• Monte Carlo (tfp.monte_carlo): Tools for computing Monte Carlo expectations.

TensorFlow Probability is under active development. Interfaces may change at any time.

Examples

See tensorflow_probability/examples/ for end-to-end examples. It includes tutorial notebooks such as:

• Linear Mixed Effects Models. A hierarchical linear model for sharing statistical strength across examples.
• Eight Schools. A hierarchical normal model for exchangeable treatment effects.
• Hierarchical Linear Models. Hierarchical linear models compared among TensorFlow Probability, R, and Stan.
• Bayesian Gaussian Mixture Models. Clustering with a probabilistic generative model.
• Probabilistic Principal Components Analysis. Dimensionality reduction with latent variables.
• Gaussian Copulas. Probability distributions for capturing dependence across random variables.
• TensorFlow Distributions: A Gentle Introduction. Introduction to TensorFlow Distributions.
• Understanding TensorFlow Distributions Shapes. How to distinguish between samples, batches, and events for arbitrarily shaped probabilistic computations.
• TensorFlow Probability Case Study: Covariance Estimation. A user's case study in applying TensorFlow Probability to estimate covariances.

It also includes example scripts such as:

• Variational Autoencoders. Representation learning with a latent code and variational inference.
• Vector-Quantized Autoencoder. Discrete representation learning with vector quantization.
• Disentangled Sequential Variational Autoencoder Disentangled representation learning over sequences with variational inference.
• Grammar Variational Autoencoder. Representation learning over productions in a context-free grammar.
• Latent Dirichlet Allocation (Distributions version, Edward2 version). Mixed membership modeling for capturing topics in a document.

• Deep Exponential Family. A deep, sparse generative model for discovering a hierarchy of topics.

• Bayesian Neural Networks. Neural networks with uncertainty over their weights.
• Bayesian Logistic Regression. Bayesian inference for binary classification.

Installation

Stable Builds

To install the latest version, run the following:

# Installing with the `--upgrade` flag ensures you'll get the latest version.
pip install --user --upgrade tensorflow-probability  # depends on tensorflow (CPU-only)

TensorFlow Probability depends on a recent stable release of TensorFlow (pip package tensorflow). See the TFP release notes for details about dependencies between TensorFlow and TensorFlow Probability.

Note: Since TensorFlow is not included as a dependency of the TensorFlow Probability package (in setup.py), you must explicitly install the TensorFlow package (tensorflow or tensorflow-gpu). This allows us to maintain one package instead of separate packages for CPU and GPU-enabled TensorFlow.

To force a Python 3-specific install, replace pip with pip3 in the above commands. For additional installation help, guidance installing prerequisites, and (optionally) setting up virtual environments, see the TensorFlow installation guide.

Nightly Builds

There are also nightly builds of TensorFlow Probability under the pip package tfp-nightly, which depends on one of tf-nightly and tf-nightly-gpu. Nightly builds include newer features, but may be less stable than the versioned releases.

Installing from Source

You can also install from source. This requires the Bazel build system.

# sudo apt-get install bazel git python-pip  # Ubuntu; others, see above links.
git clone https://github.com/tensorflow/probability.git
cd probability
bazel build --copt=-O3 --copt=-march=native :pip_pkg
PKGDIR=$(mktemp -d)
./bazel-bin/pip_pkg $PKGDIR
pip install --user --upgrade $PKGDIR/*.whl

Community

As part of TensorFlow, we're committed to fostering an open and welcoming environment.

• Stack Overflow: Ask or answer technical questions.
• GitHub: Report bugs or make feature requests.
• TensorFlow Blog: Stay up to date on content from the TensorFlow team and best articles from the community.
• Youtube Channel: Follow TensorFlow shows.
• [email protected]: Open mailing list for discussion and questions.

See the TensorFlow Community page for more details. Check out our latest publicity here:

• Coffee with a Googler: Probabilistic Machine Learning in TensorFlow
• Introducing TensorFlow Probability

Contributing

We're eager to collaborate with you! See CONTRIBUTING.md for a guide on how to contribute. This project adheres to TensorFlow's code of conduct. By participating, you are expected to uphold this code.

References

If you use TensorFlow Probability in a paper, please cite:

• TensorFlow Distributions. Joshua V. Dillon, Ian Langmore, Dustin Tran, Eugene Brevdo, Srinivas Vasudevan, Dave Moore, Brian Patton, Alex Alemi, Matt Hoffman, Rif A. Saurous. arXiv preprint arXiv:1711.10604, 2017.

(We're aware there's a lot more to TensorFlow Probability than Distributions, but the Distributions paper lays out our vision and is a fine thing to cite for now.)