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 (
tf.contrib.distributions
,tf.distributions
): A large collection of probability distributions and related statistics with batch and broadcasting semantics. - Bijectors (
tf.contrib.distributions.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. - 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.
Installation
To install the latest version, run the following:
pip install --user --upgrade tfp-nightly # depends on tensorflow (CPU-only)
TensorFlow Probability depends on a current nightly release of TensorFlow
(tf-nightly
); the --upgrade
flag ensures you'll automatically get the latest
version.
We also provide a GPU-enabled package:
pip install --user --upgrade tfp-nightly-gpu # depends on tensorflow-gpu (GPU enabled)
Currently, TensorFlow Probability does not contain any GPU-specific code. The
primary difference between these packages is that tensorflow-probability-gpu
depends on a GPU-enabled version of 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.
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 --config=opt --copt=-O3 --copt=-march=native :pip_pkg
PKGDIR=$(mktemp -d)
./bazel-bin/pip_pkg $PKGDIR
pip install --user --upgrade $PKGDIR/*.whl
Usage
Access the library using
import tensorflow_probability as tfp
Examples
See tfp.examples
for end-to-end examples. It includes tutorial notebooks such as:
Example scripts can also be run under command line. For example,
python -m tensorflow_probability.examples.vae
trains a variational auto-encoder to generate MNIST digits.
Contributing
We're eager to collaborate with you! Feel free to open an issue on
GitHub and/or send us your
pull requests. See CONTRIBUTING.md
for more details.
This project adheres to TensorFlow's code of conduct. By
participating, you are expected to uphold this code.
References
- 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.