This application provides the basic steps of a machine learning workflow from a graphical user interface built with Shiny. It uses the functionalities of the R-package mlr3.
Current functionalities of mlr3shiny are:
- Data import
- Creation of a task for supervised learning (regression, classification)
- Use of a set of algorithms as learners
- Training and evaluation of the generated models
- Benchmarking to compare several learners on a task simultaneously
- Prediction on new data using the trained learner
- Explain trained learner
Tetzlaff, L. and Szepannek G. (2022): mlr3shiny—State-of-the-art machine learning made easy, SoftwareX 20, DOI: 10.1016/j.softx.2022.101246.
Install the package in R via CRAN:
install.packages(mlr3shiny)Install the development version of the package in R from GitHub.
remotes::install_github("https://github.com/LamaTe/mlr3shiny.git")Launch the application via:
mlr3shiny::launchMlr3Shiny()Navigate over the different steps of the workflow using the menu bar. The tabs are chronologically ordered. The question mark in the top-right corner provides more information on the functionalities and purpose of each section. Start by importing a dataset. Then define a task (the problem to be solved) in the 'task' tab. Example tasks are already provided. Select different learners (algorithms) in the 'learner' tab and train and evaluate a model in 'train & evaluate'. Resampling strategies can be applied in a sub-section of 'train & evaluate'. Alternatively, different learners can be compared in a benchmark. Use the final model to make a prediction on new data in the 'predict' tab. An explanation of the final model from the predict tab can be made in the 'explain' tab.