Functionality to skip writing of .hdf5 object with input data

[davidwalter2]

The current workflow is to create the input data file via the tensorwriter and then load it as an inputdata object.
It would be nice to have the possibility to directly create the inputdata object without the need to write the object.