CSVMixin does not read missing values as NaN
We do not explicitly read in floats, but set dtype=None
to let NumPy guess the type. This guessing is needed because the first column is (sometimes) a string representation of datetimes, and to support using a comma as a decimal separator.
If a column is entirely empty in the input file, NumPy seems to guess that it's a boolean (full of False). This is then forcibly converted to a float array when storing it in a Timeseries, causing it to be a bunch of zeros.
It would be both faster and would solve this issue if we make sure that all columns are read as floats, except for the first column if with_time
is set.
Edited by Tjerk Vreeken