greta.distributions

Background

Explain the motivation for your coding project. What problem would it solve?

greta is an R package for statistical modelling, using Google's Tensorflow as its underlying engine. The greta package has many distributions already built into the package, but there is often a need for more custom, less commonly used distributions. greta.distributions is an R package that provides extra probability distributions to be used with greta, as well as tools for writing new distributions.

Probability distributions are essential component of the foundation for

Related work

Although there are many extra distribution packages in the R programming language, such as distributional, extraDistr, glmmTMB, gamlss.dist, and gamlss.tr, and more listed at the CRAN Task View for probabity distributions. The point is however that in order for distributions to be used within greta, they must be written out using TensorFlow.

Details of your coding project

The goal of this project is to contribute new distributions to the greta.distributions package, add documentation, tests, and a vignette on how to create new distributions. Specifically we would like the following distributions implemented:

• zero inflated negative binomial distribution
• zero inflated poisson distribution
• conditional bernoulli
• multivariate probit
• discrete lognormal and discrete normal
• tweedie distribution

As well as developing tools in the R package to assist users in writing their own greta distribution. Specifically,

• helpers for creating new distributions
• Helpers for creating tests for new distributions
• Writing a vignette on creating a new greta distribution

Expected impact

There is an increasing need for the distributions listed above in the field of ecology. By making these distributions accessible we will be

Mentors

Please get in touch with EVALUATING mentor Nicholas Tierney <[email protected]> and Nick Golding <[email protected]> after completing at least one of the tests below.

Tests

Do one, or more, if you want! If you do more hard tests you are more likely to be selected.

• Easy: download and install both greta and greta.distributions. Run the "common examples" examples at: https://greta-stats.org/articles/example_models#common-models.
• Medium: Check out https://github.com/greta-dev/greta.distributions/pull/16 and run tests on this - see if you can get tests to pass?
• Hard: write out a Tweedie distribution in R, see https://github.com/greta-dev/greta.distributions/issues/18 for details. Implement rtweedie() for sampling from tweedie, and dtweedie for evaluating the density at a given location.

Solutions of tests

Contributors, please post a link to your test results here.

• contributor name, link to github repository