Study level

  • Vacation research experience scheme

Faculty/School

Topic status

We're looking for students to study this topic.

Research centre

Supervisors

Mr Joshua Bon
Position
Research Fellow in Sequential Monte Carlo Methods
Division / Faculty
Faculty of Science

Overview

Bayesian additive regression trees (BART) are semiparametric models often used in machine learning tasks and causal inference. Currently there is a huge interest in using these models for statistical inference and prediction.

The tidytreatment R package is designed to facilitate using BARTs from several different R packages in a consistent format. The package contains tools to visualise and investigate these BARTs to assist practitioners in developing reproducible analyses that can interface with other "tidy" packages (e.g. tidyverse packages and tidybayes).

Tidytreatment aims to demystify an advanced machine learning technique and provide important methods in the model building and testing process.

Research activities

Possible research activities are broadly related to research software engineering, developing new visualisations or algorithms, and Bayesian inference. For example, a student may work to:

  • discover and implement new visualisations for BART models
  • develop new model checking methods
  • integrate tidytreatment with other packages: e.g. bcf and bartCause
  • develop new algorithms for fitting BART models to complex data
  • other tasks.

Outcomes

In close collaboration with the supervisor it is expected that a student would make progress towards some of the following goals:

  • advance the capabilities of tidytreatment: e.g. visualisation, integration, testing
  • contribute to helping researchers use BART models with confidence
  • experience developing R packages
  • introduction to research software engineering.

Skills and experience

Essential:

  • some experience coding in R (or Python, Julia, Matlab etc and willing to work in R)
  • basic understanding of statistical modelling or machine learning.

Desired (but not necessary):

  • experience with Bayesian modelling and/or computation
  • some experience writing R packages
  • some experience with ggplot2 and the tidyverse.

Keywords

Contact

Contact the supervisor for more information.