Work with Model Predictions and Track Model Performance
Creating and visualizing model predictions takes advantage of many different types of R content and the ability to deploy them on RStudio Connect.
Predictions can be made available in a web application and to other services via an API
Continuous monitoring of the predictions can be used to watch for model drift and changes in goodness-of-fit over time
The following model deployments in RStudio Connect are from the Bike Prediction example.
The bike prediction model outputs data in two ways:
The model quality assessment script bulk loads predictions into a database
The Plumber API loads the model and provides live predictions based on the newest data
Depending on the relative frequency of model retraining and input data refreshes, either of these can be a good way to serve model data.
The model performance app allows for comparisons over time between models:
Full Example: https://solutions.rstudio.com/tour/bike_predict/
Git Repository: https://github.com/sol-eng/bike_predict