predx: An R package for predictions

predx aims to be a flexible, lightweight framework for working with predictions. The package implements a standardized embedded data frame in R which captures a variety of prediction types. Predictions can be imported from or exported to CSV or JSON files, including a predx-specific JSON format that drastically reduces file sizes. This format can be particularly helpful for transfering large sets of forecasts.

Currently, the package supports basic manipulation and verification of predictions, including those for Epidemic Prediction Initiative challenges. The FluSight Vignette and Aedes Vignette demonstrate how the package can be used to validate forecasts for these challenges. Code is also provided to interface with the FluSight R package for scoring FluSight forecasts (see the FluSight Vignette).

Four predx classes (S4) are now available: Point, Binary, BinLwr, and BinCat. More will be added in the future.

To get started, load the devtools package and install predx from this repository with the vignettes. The documentation is currently quite sparse, but the vignettes provide a guide to basic functions.

library(devtools)
devtools::install_github('cdcepi/predx', build_vignettes = TRUE)

If you find bugs, please open an issue.