parallelize or impute just one sample

[nsmackler]

Would it be possible to either parallelize the training (as mentioned in your boostme code " # TODO: add in parallel option for this instead of loop (?)" )

Or, perhaps to just add in an argument to only impute 1 sample at a time? then you could impute by distributing jobs across a cluster to impute 1 sample at a time.

Thanks!