基于训练集的测试集归因

Question

样本数据

#creates a data frame with two features and 100 observations
train = data.frame(salary = sample(2000:3000, 100, replace = TRUE),
                 revenue = sample(c("A", "B"), 100, replace = TRUE))

#five NAs on random positions for the salary feature
train[sample(1:nrow(train), 5), "salary"] <- NA

test = data.frame(salary = sample(2000:3000, 100, replace = TRUE),
                   revenue = sample(c("A", "B"), 100, replace = TRUE))

test[sample(1:nrow(test), 5), "salary"] <- NA

数据转换

transformation <- preProcess(train[, ], method = c("medianImpute", 
"scale", "center"))


transformed_train <- predict(transformation, newdata = train[, ])

transformed_test <- predict(transformation, newdata = test[, ])

交叉验证

trControl = trainControl(method = "repeatedcv", 
                         number = 5, 
                         repeats = 5, 
                         classProbs = TRUE,
                         savePredictions = TRUE, 
                         verboseIter = TRUE)

建模

model_knn <- caret::train(revenue ~ .,
                          data = transformed_train,
                          method = "knn",
                          preProcess = c("scale", "center", "knnImpute"),
                          trControl = trControl)

confusionMatrix(predict(model_knn, transformed_test), transformed_test$revenue)

Cotext

以上，我已经复制了一个示例，说明如何使用Caret程序包中的preProcess函数来估算数据。 PreProcess也可以在建模阶段使用。

问题

我不想使用Caret的预处理功能，而是想知道如何实现MICE插补包？这个想法是在测试集上使用推算训练集的结果。

我想唯一需要更新的是数据转换部分。