JuiceBoxCV: JuiceBoxCV

In sjoshistrats/JuiceBox: Caret Extensions

Description

Performs cross validation with a given pipeline

Usage

JuiceBoxCV(X_train, Y_train, numFolds, numRepeats, parCV, numCores, seedNum,
  verbose_p, fn, fn_params)

Arguments

X_train	Training Data (excludes the response/target we wish to predict ) that will be fed into the pipeline function.
Y_train	Training Response/Target - The response/target that will be fed into the pipeline function.
numFolds	Integer indicating the number of folds to use in the cross validation procedure.
numRepeats	Integer indicating the number of times to repeat cross validation with numFolds.
parCV	Boolean indicating whether to parallelize the training prodcedure.
numCores	Integer indicating the number of cores to use.
seedNum	Integer indicating the seed number. Using the same seed will generate the same folds.
verbose_p	Boolean indicating if cross validation details should be printed out the screen.
fn	The pipeline function. The pipeline function must take parameters training data, training response, validation data, validation response. See examples for details.
fn_params	Additional parameters to supply to the pipeline function. See examples for details.

Value

Average cross validation score across all the folds and repeats.

Examples

library(JuiceBox)
library(Metrics)
library(xgboost)

# Toy data set for classification
irisAllMat <- iris
irisTrainMat <- irisAllMat[,c(1:4)]
irisTrainResponse <- irisAllMat[,c(ncol(irisAllMat))]
irisTrainResponse <- factor(ifelse(irisTrainResponse == "setosa", "Yes", "No"))

# Toy data set for regression
mtcarsMat <- mtcars
mtcarsResponse <-mtcarsMat[,1]
mtcarsMat <- mtcarsMat[,c(2:ncol(mtcarsMat))]

# Pipelines
xgbPipeline_regression <- function(X_train, Y_train, X_test, Y_test, params)
{
  Y_train <- as.numeric(Y_train) - 1
  xgbFit <- xgboost(data = as.matrix(X_train), label = Y_train, max.depth = params[1],
                    eta = params[2], nround = params[3], objective = "reg:linear", verbose = 1)
  predictions <- predict(xgbFit, newdata = as.matrix(X_test))
  actual <- as.numeric(Y_test) - 1
  rmseValue <- rmse(predictions, actual)
  return(-rmseValue)
}

xgbPipeline_classification <- function(X_train, Y_train, X_test, Y_test, params)
{
  Y_train <- as.numeric(Y_train) - 1
  xgbFit <- xgboost(data = as.matrix(X_train), label = Y_train, max.depth = params[1],
                    eta = params[2], nround = params[3], objective = "binary:logistic", verbose = 1)
  predictions <- predict(xgbFit, newdata = as.matrix(X_test))
  actual <- as.numeric(Y_test) - 1
  logLossValue <- logLoss(actual, predictions)
  return(-logLossValue)
}

xgbPipeline_regression_extraction <- function(X_train, Y_train, X_test, Y_test, params)
{
  Y_train <- as.numeric(Y_train) - 1
  xgbFit <- xgboost(data = as.matrix(X_train), label = Y_train, max.depth = params[1],
                    eta = params[2], nround = params[3], objective = "reg:linear", verbose = 1)
  predictions <- predict(xgbFit, newdata = as.matrix(X_test))
  return(predictions)
}

xgbPipeline_classification_extraction <- function(X_train, Y_train, X_test, Y_test, params)
{
  Y_train <- as.numeric(Y_train) - 1
  xgbFit <- xgboost(data = as.matrix(X_train), label = Y_train, max.depth = params[1],
                    eta = params[2], nround = params[3], objective = "binary:logistic", verbose = 1)
  predictions <- predict(xgbFit, newdata = as.matrix(X_test))
  return(predictions)
}

# Testing JuiceBoxCV
JuiceBoxCV(X_train = irisTrainMat, Y_train = irisTrainResponse, numFolds = 2,
           numRepeats = 2, parCV = FALSE, numCores = 8, seedNum = 101,
           fn = xgbPipeline_classification, fn_params = c(3, 2, 5), verbose_p = 1)

JuiceBoxCV(X_train = mtcarsMat, Y_train = mtcarsResponse, numFolds = 2,
           numRepeats = 2, parCV = FALSE, numCores = 8, seedNum = 101,
           fn = xgbPipeline_regression, fn_params = c(3, 2, 10), verbose_p = 1)

sjoshistrats/JuiceBox documentation built on May 30, 2019, 12:05 a.m.