Hi!
Currently in DLMReader/InMemoryDatasets, Floats very close to zero (under 5e-324) are parsed as missing values. It seems to me that the prefered behaviour should be to parse those as 0.0 (like CSV/DataFrames does) in order to distinguish from "real" missing values. Keeping the warning (as opposed to CSV) is probably a good idea though. What do you think about it?

Thanks a lot!