sgmcmc implements popular stochastic gradient Markov chain Monte Carlo (SGMCMC) methods including stochastic gradient Langevin dynamics (SGLD), stochastic gradient Hamiltonian Monte Carlo (SGHMC) and stochastic gradient Nosé-Hoover thermostat (SGNHT). The package uses automatic differentiation, so all the differentiation needed for the methods is calculated automatically. Control variate methods can be used in order to improve the efficiency of the methods as proposed in the recent publication.

The package is built on top of the TensorFlow library for R, which has a lot of support for statistical distributions and operations, which allows a large class of posteriors to be built. More details can be found at the TensorFlow R library webpage, also see the TensorFlow API for full documentation.

To cite the sgmcmc package, please reference the accompanying paper. Sample Bibtex is given below:

@article{sgmcmc-package,
  title={sgmcmc: An R package for stochastic gradient Markov chain Monte Carlo},
  author={Baker, Jack and Fearnhead, Paul and Fox, Emily B and Nemeth, Christopher},
  journal={Journal of Statistical Software},
  volume={91},
  number={3},
  pages={1--27},
  year={2019}
}

sgmcmc requires TensorFlow for R to be installed, which requires packages that can't be automatically built by R, so has a few steps:

• Install the sgmcmc R package: install.packages("sgmcmc").
• Install the required python packages (including TensorFlow and TensorFlow Probability) by running: sgmcmc::installTF().

If you already have the TensorFlow and TensorFlow probability packages installed, then this should be autodetected by the package and you can skip the final step. Make sure these are up to date though, as the TensorFlow API is under active development and still changes quite regularly. Especially ensure that your TensorFlow and TensorFlow probability modules are compatible.

It's recommended you start here. This getting started page outlines the general structure of the package and its usage.

There's also worked examples for the following models (these will be extended as the package matures):

• Multivariate Gaussian
• Gaussian Mixture
• Logistic Regression

The SGMCMC algorithms can also be run step by step, which allows custom storage of parameters using test functions, or sequential estimates. Useful if your chain is too large to fit into memory! This requires a better knowledge of TensorFlow. An example of this is given in the neural network vignette.

Finally full details of the API can be found here.

For the source code, and bug reporting, see the Github page.

If you are having issues running the examples, as a first port of call please make sure your TensorFlow installation is the most up to date version. A lot of issues are simply because the TensorFlow API has changed. If you're still having issues, please file a bug report.