A project to demonstrate automated Shiny app deployment via GitHub Actions and RStudio Connect (RSC). The app is deployed at https://colorado.rstudio.com/rsc/uw-dashboard.
The current content of the dashboard consists of a visualization tool to analyze additive variable attributions in a personal automobile insurance loss cost model. The model is built using the R interface to TensorFlow/Keras and model interpretation is powered by the DALEX package. For details, refer to the paper below and the associated repo.
Kevin Kuo and Daniel Lupton. Towards Explainability of Machine Learning Models in Insurance Pricing. arXiv:2003.10674, 2020.
Changes to the app are proposed via pull requests (PRs). When a new PR is opened,
the code is bundled then deployed as a new temporary app on RSC. At this point,
user testing can be done to ensure that the app is behaving as expected. Upon approval,
one can merge the PR to master, which triggers a deployment to the persistent
production endpoint.
Features requests, bug reports, and PRs are all welcome!