mlr fits and condvis2

mlr is a R package that offers a unified interface to machine learning models. By writing an interface between condvis2 and mlr a vast number of machine learning fits may be explored with condvis. Presently, regression, classification and clustering varieties of mlr learners work with `condvis’.

A list of models supported by mlr is found on this link: https://mlr.mlr-org.com/articles/tutorial/integrated_learners.html

Regression

Set up the task, learner and train the model.

library(mlr)
library(MASS)
library(condvis2)
Boston1 <- Boston[,9:14]

rtask <- makeRegrTask(id = "bh", data = Boston1, target = "medv")
rmod <- train(makeLearner("regr.lm"), rtask)
rmod1 <- train(makeLearner("regr.fnn"), rtask)

Use condvis to explore the models:

condvis(Boston1, model=list(rmod,rmod1), response="medv", sectionvars="lstat")

Choose tour “Diff fits” to explore differences between the fits

Some tasks, for example linear regression, support standard errors and so confidence intervals. This option needs to be added to makeLearner. Then, tell condvis to plot an interval using pinterval="confidence for that fit.

rmod <- train(makeLearner("regr.lm", predict.type="se"), rtask)
condvis(Boston1, model=rmod, response="medv", sectionvars="lstat", predictArgs=list(list(pinterval="confidence")))

Classification

Set up the task, learner and train the model.

cltask = makeClassifTask(data = iris, target = "Species")
cllrn = makeLearner("classif.lda",predict.type = "prob") # need predict.type ="probs" to get probs
clmod = train(cllrn, cltask)

Explore with condvis:

condvis(iris, model=clmod, response="Species", sectionvars=c("Petal.Length", "Petal.Width"), pointColor="Species")

Click on “Show probs” to see class probabilities.

Clustering


ctask = makeClusterTask(data = iris[,-5])
clrn = makeLearner("cluster.kmeans") 
cmod = train(clrn, ctask)

Add the predicted class to the data to act as the response:

library(dplyr)
iris1 <- iris

iris1$pclass <- cmod %>%
  predict(newdata=iris[,-5]) %>%
  getPredictionResponse() %>% 
    as.factor()

condvis(data = iris1, model = cmod, 
        response="pclass",
        sectionvars=c("Petal.Length", "Petal.Width"), 
        conditionvars=c("Sepal.Length", "Sepal.Width"),pointColor="Species"
)

Catherine B. Hurley

2022-09-13

Regression

Classification

Clustering