Update on gradient boosting R Shiny app
Dredging up an old R Shiny web application that I got started on but didn’t get very far with. The app allows the user to perform stochastic gradient boosting using the R gbm package.
Initially I halted progress on it because I encountered an error (see previous post for details) that I couldn’t find the cause of with certainty. Recently I discovered that others had a similar problem and it was a bug in the gbm package. It did not have anything to do with the Shiny package as I had expected. It was just hard to find the exact conditions for replicating the error. Anyway, at the time of this writing, the gbm package maintainer has fixed the bug in the development version of the gbm package (the version on CRAN still has the bug). Since switching to the development version I’ve seen the problem is now resolved.
This is great because now there is some point to finding time to continue developing this app with additional features. It currently only provides output plots and tables of relative influence and different types of estimated prediction error. The only other change I have implemented recently is a switch to display tables in a prettier and interactive fashion using the Shiny package functions
dataTableOutput, in place of
tableOutput respectively, which were not available when I first began working on the app.