Better way of handling missing data.

[xehu]

The repository, and the TPM work as a whole, currently contain a number of naive ways to handle missing data. This Issue proposes to handle them more thoughtfully.

e.g., from the multi-task data cleaning:

	threshold = 0.8  # 80% threshold
	user_info = user_info.dropna(axis=1, thresh=int(threshold * len(user_info)))
	#### TODO --- figure out better way to handle NA's here!

from the HorseRaceDataSet constructor:

  if(self.fillna_options == 'mean'):
            # Fill NA with mean -- TODO: add other options!
            data.fillna(data.mean(numeric_only=True), inplace=True)