Impute Missing Values with Caret

Question

I am working on the Kaggle Titanic competition and I have a question regarding imputing missing values. I am trying to use the Caret package and my training set consists of factors as well as numbers.

I want to use the preProcess function in Caret to impute the missing values, but before using preProcess, I need to convert all my factors into dummy variables with the dummyVars function.

dummies  = dummyVars(survived ~ . -1, data = train, na.action = na.pass)
xtrain = predict(dummies, train)

However, in the process of using dummyVars to convert the factors, all the NAs are predicted by some unknown algorithm and the missing age columns all become 1's even though I have specified na.action = na.pass. I want to convert my factors into dummy variables WITHOUT having the NAs touched so I can use then use the preProcess function to impute them. How can I do this?

Thank you.

dput here:

structure(list(survived = structure(c(1L, 2L, 2L, 2L, 1L, 1L, 
1L, 1L, 2L, 2L, 2L, 2L, 1L, 1L, 1L, 2L, 1L, 2L, 1L, 2L), .Label = c("0", 
"1"), class = "factor"), pclass = structure(c(3L, 1L, 3L, 1L, 
3L, 3L, 1L, 3L, 3L, 2L, 3L, 1L, 3L, 3L, 3L, 2L, 3L, 2L, 3L, 3L
), .Label = c("1", "2", "3"), class = "factor"), sex = structure(c(2L, 
1L, 1L, 1L, 2L, 2L, 2L, 2L, 1L, 1L, 1L, 1L, 2L, 2L, 1L, 1L, 2L, 
2L, 1L, 1L), .Label = c("female", "male"), class = "factor"), 
    age = c(22, 38, 26, 35, 35, NA, 54, 2, 27, 14, 4, 58, 20, 
    39, 14, 55, 2, NA, 31, NA), sibsp = c(1, 1, 0, 1, 0, 0, 0, 
    3, 0, 1, 1, 0, 0, 1, 0, 0, 4, 0, 1, 0), parch = c(0, 0, 0, 
    0, 0, 0, 0, 1, 2, 0, 1, 0, 0, 5, 0, 0, 1, 0, 0, 0), fare = c(7.25, 
    71.2833, 7.925, 53.1, 8.05, 8.4583, 51.8625, 21.075, 11.1333, 
    30.0708, 16.7, 26.55, 8.05, 31.275, 7.8542, 16, 29.125, 13, 
    18, 7.225), embarked = structure(c(4L, 2L, 4L, 4L, 4L, 3L, 
    4L, 4L, 4L, 2L, 4L, 4L, 4L, 4L, 4L, 4L, 3L, 4L, 4L, 2L), .Label = c("", 
    "C", "Q", "S"), class = "factor")), .Names = c("survived", 
"pclass", "sex", "age", "sibsp", "parch", "fare", "embarked"), row.names = c(NA, 
20L), class = "data.frame")

Answer 1

This first part is a bug; the NA values should not be 1's (obviously). In the meantime, you can use model.matrix to generate the dummy variables, but you might have to do this at once for all of the data. Also, if you are using train, you can use the formula method. Overall, that is a better approach.

I'll fix this in the next few weeks. I'm about to release a version of caret and this, plus UseR, will delay me a bit.

EDIT: a new version will be released in the next week that fixes the bug

Max