The NSW Branch of the SSA is offering this two day workshop.
Often, statistical analyses require custom models that cannot be fitted using off-the shelf statistical software, but can be estimated by MCMC by specifying the model in specialised software, the most popular of which are BUGS, JAGS and Stan. Greta is a package for statistical modelling in R that has three core differences to these available alternatives:
1. simple: greta models are written right in R, so there's no need to learn another language like BUGS or Stan
2. scalable: greta uses Google TensorFlow so it's fast even on massive datasets, and runs on CPU clusters and GPUs
3. extensible: it's easy to write your own R functions and packages using greta
We will start with simple linear models on real ecological data, and gradually expand the models to be more complex and better represent the data. We will also have time at the end of the course to discuss fitting models specific to your own work - so feel free to bring along a problem you’d like to discuss!
Presenter:
Professor Nick Golding is an infectious disease modeller with a focus on globally-important pathogens. His work combines mathematical and statistical modelling, ecology, public health, and research software engineering. Since completing a PhD on mosquito ecology, he has developed models and maps of the risk posed by some of the world’s most important and neglected diseases – including malaria, Dengue fever, Chikungunya, Ebola, and COVID-19. He has a preference for semi-mechanistic Bayesian models applied to large, noisy datasets and developed the greta R package to handle both the scale and uniqueness of these types of models.
Dr. Nick Tierney completed his undergrad and honours in Psychological Science, then took an unconventional turn into a PhD in Statistics. He now works as a research software engineer with Dr. Nick Golding at the Telethon Kids Institute. He is currently working on improving and maintaining the greta (https://greta-stats.org/) R package for statistical modelling. He is also interested in implementing workflows to automate data analysis. Dr. Tierney’s research interests are broad, but centred around improving data analysis. This includes exploratory data analysis, statistical modelling, diagnostics, and understanding how colour choice can impact decision making. Dr. Tierney is a strong believer in free and open source software, and has written several popular R packages to improve data analysis, which can be seen on his software page: http://njtierney.com/software.
Requirements:
This course is designed for those who want to learn how to do Bayesian modelling using the greta software. We assume users have the following background/experience:
- Familiarity with R
- Experience using linear models
A rudimentary understanding of Bayesian inference
We recommend trying to install the software from github on your laptop/machine ahead of time with remotes::install.github("greta-dev/greta") but we will also provide virtual environments to use in the course. Course materials will be made available online.
After this course you will be able to:
- Fit and predict from Bayesian generalised linear models in greta
- Check model convergence and fit (including prior and posterior predictive checks)
- Summarise MCMC outputs
- Be able to fit more advanced models including mixture and hierarchical models
- Create visualisations and tables of the model outputs for use in understanding model fit and for publication.
The Early Bird deadline closes on 12 October 2021.
Cancellation Policy:
Cancellations received prior to Thursday, November 11, will be refunded, minus a $20 administration fee. From then onward no part of the registration fee will be refunded. However, registrations are transferable within the same organisation. Please advise any changes to events@statsoc.org.au.
@StatSocAus
