Automation Machine Learning

autoML.R

library(mlr)
library("parallelMap")

# install all packages
install.packages(unlist(strsplit(as.data.frame(listLearners())$package,",")))

# Load data
dataTrain <- NULL
dataTest <- NULL
dataTarget <- "response"

parallelStartSocket(parallel::detectCores())
# Create Task
trainTask <- makeClassifTask(data = dataTrain, target = dataTarget)
testTask <- makeClassifTask(data = dataTest, target = dataTarget)

# Loop
learners <- makeModelMultiplexer(lapply(names(presetML), makeLearner))

#
params <- makeModelMultiplexerParamSet(learners, presetML)

#tuneParamsMultiCrit
tunes <- tuneParams(learner = learners, task = trainTask, resampling = makeResampleDesc("CV",iters = 10L),  par.set = params, control = makeTuneControlIrace(maxExperiments = 500L), measures = acc)

#
learners <- makeModelMultiplexer(lapply(names(presetML), setHyperPars, par.vals = tunes))

#
models <- train(learners, trainTask)

#
predicts <- predict(learners, testTask)

#
parallelStop()

presetML.R

presetML = list(
  # Extreme Gradient Boosting
  classif.xgboost = makeParamSet(
    makeIntegerParam(id = "nrounds", lower = 200, upper = 600, default = 300),
    makeIntegerParam(id = "max_depth", lower = 3, upper = 20, default = 10),
    makeNumericParam(id = "lambda", lower = 0.55, upper = 0.60, default = 0.60),
    makeNumericParam(id = "eta", lower = 0.001, upper = 0.5, default = 0.1),
    makeNumericParam(id = "subsample", lower = 0.10, upper = 0.80, default = 0.5),
    makeNumericParam(id = "min_child_weight", lower = 1, upper = 5, default = 4),
    makeNumericParam(id = "colsample_bytree", lower = 0.2, upper = 0.8, default = 0.5)
  ),
  #rpart
  classif.rpart = makeParamSet(
    makeIntegerParam("minsplit",lower = 10, upper = 50, default = 30),
    makeIntegerParam("minbucket", lower = 5, upper = 50, default = 20),
    makeNumericParam("cp", lower = 0.001, upper = 0.2, default = 0.1)
  )

)