Could starting frames use 'nan' instead of '-'

[asgibson]

A starting frame of data is filled with '-' in place of all values. As data comes in they are replaced with the value for that specific data point; however, this means we are mixing strings and floats in our data frame. We could instead use float('nan') as our initial values. Would that be appropriate without causing errors?

[dragonejt]

I think this is a good argument to move towards using numpy arrays for frames everywhere instead of Python lists. Simple numpy arrays can are of a single type and precision, and we can use the special value np.nan. This would require a bit more work though.