writeARFF: Only quote strings, not factors or numbers

[mboecker]

This encodes only columns with type "string" with quotation marks.

@jakob-r

[mboecker]

We could also do something like this, if it is preferred:

  if (quote == "all") {

    # Quote everything
    quote = TRUE
  } else if(quote == "fast") {

    # find out which columns to quote, we currently quote strings but not factors
    quote = which(sapply(colnames(x), function(colname) { is.character(x[,colname]) }))
  } else if(quote == "slow") {

    # quote strings and factors, which contain whitespace (as per specification)
    quote = which(sapply(colnames(x), function(colname) { any(sapply(x[,colname], function(str) grepl(pattern = "\\s+", str))) }))
  } else {

    stop("writeARFF: missing argument quote. please specify all, fast or slow")
  }

[jakob-r]

why so comlplicated?

vapply(data, is.character, logical(1))

but actually i would like something like this

vapply(data, function(x) is.factor(x) && all(!is.na(as.numeric(levels(x)))), logical(1))

Background info for the others:
If we have a column that has e.g. 0,1 and is our label. The MOA Framework did not classify correctly if they where enquoted as '0' and '1'. This can be a bug on their side as well but it seems that many datasets dont store factors enquoted so i wanted this option.

[mboecker]

Hey @jakob-r
I adapted your code, but added is.numeric as a case to not use quotes.
This works nicely and results in strings and factors that contain non-numeric levels being quoted.