modelDown generates a website with HTML summaries for predictive models. Is uses DALEX explainers to compute and plot summaries of how given models behave. We can see how well models behave (Model Performance, Auditor), how much each variable contributes to predictions (Variable Response) and which variables are the most important for a given model (Variable Importance). We can also compare Concept Drift for pairs of models (Drifter). Additionally, data available on the website can be easily recreated in current R session (using
pkgdown documentation: https://mi2datalab.github.io/modelDown/
An example website for regression models: https://mi2datalab.github.io/modelDown_example/
Do you want to start right now ? Check out our getting started guide.
Index page presents basic information about data provided in explainers. You can also see types of all explainers given as parameters. Additionally, summary statistics are available for numerical variables. For categorical variables, tables with frequencies of factor levels are presented.
drifter package are displayed in this tab. In order to see the comparision charts, you have to provide pair of explainers as parameters (for example:
Output of function
variable_importance is presented in form of a plot as well as a table.
For each variable, plot is created by using function
variable_response. Plots can be easily navigated using links on the left side. One can provide names of variables to include in the module with argument
vr.vars (if argument is not used, plots for all variables of first explainer are generated).
In each tab you can find links with R commands. If you execute them, you can load relevant objects into current R session (
archivist package is necessary). By default data is stored and loaded from local repository. If you wish to store data on GitHub repository, please provide argument
remote_repository_path. After generating modelDown website,
repository folder must be placed under this path.