Add handling for character & factor y

[tpq]

2-part bug report / question,

-- When using character, error is thrown:

`Error in y * 0 : non-numeric argument to binary operator'

-- Factor works, but unsure whether factor is handled as discrete classes (via a logistic / multinomial regression) or coerced into continuous numeric (via linear regression)

-- Proposed solution:

Always coerce character into factor

if(class(y) == "character"){
  y <- factor(y)
}

Then convert factor into 1-hot encoding of class labels

if(class(y) == "factor"){
  keras::to_categorical(as.numeric(y)-1) # <- -1 because keras, being Python, uses zero indexing
}

[egr95]

Good catch, thanks!