All Tools
T
DataFreeOpen Source
TENSORFLOW PROBABILITY
Bayesian layers, MCMC, and VI on TensorFlow
Apache-2.0
ABOUT
Bayesian deep learning needs distributions, MCMC, and variational layers that compose with TensorFlow graphs. Ad hoc NumPy samplers sit outside TF. TensorFlow Probability ships those primitives so models train uncertainty on GPU.
INSTALL
pip install tensorflow-probabilityINTEGRATION GUIDE
1. Place a TFP distribution layer on a TensorFlow network head
2. Run Hamiltonian Monte Carlo on a TensorFlow joint density
3. Train a variational autoencoder with TFP bijectors
4. Sample a Bayesian neural net without leaving the TF graph
TAGS
pythontensorflowprobabilisticbayesianmcmcvariational-inference