iml–summary

This dashboard provides an iml-summary for your data. The app shows the partial dependence plots and permutation feature important for any machine learning model. It can also explain individual predictions with Shapley Values.

File Structure

The general structure of this repo does have the following tree :

├───app 
├───pdPlots
├───example
│   └───PrediObj
│   └───test_model
├───www
|└─── Cover
|
└───libraries

How to use

You can use this shinydashboard app for your predictor object (R6Class object), which should hold any machine learning model (mlr, caret, randomForest, …) and the data to be used of analysing the model. The package iml and mlr can help you. There is an example of R code how to produce an .RDS model object with iml and mlr as below.

There is a right-click menu in this shinydashboard. Click Global Effects, save your predictor object as .RDS file and upload it. Then, you get a table in few minutes, which includes variables’ names, corresponding value ranges, partial dependence plots(PDPs) and their feature importance values. Click Local Interpretation, you will see the data set on the top. Then, if you select a certain row, the seed and the number of Monte Carlo samples for estimating the shapley value, the corresponding shapley values and its plot will be shown automatically as follows.

Read more about the methods in the Interpretable Machine Learning book

Read more about the information in the Machine Learning in R

Example

• First we train a Tree-Based Regression Models to predict the Boston median housing value. Open R file test_model.R, run the code as follows, the pred.RDS will be saved in the file \example

rm(list = ls(all.names = T))
library(iml)
library(mlr)
lrn = makeLearner("regr.rpart")
tsk = bh.task
dat = getTaskData(bh.task)
mod = train(lrn, tsk)
pred = Predictor$new(mod, dat)
saveRDS(pred, file = "example/pred.RDS")

• Open R file app.R, then, click the button “run App”, which is placed right above. And you will get a shinydashboard. Click Global Effects on the right-click menu, upload the pred.RDS……(see above)