This PR modifies how SnowStorm values are read in: they are now read in via the SnowstormParametersDict rather than a list of variables that are required to be specified in the correct order. A feature for an arbitrary label key name is also added to the data module, in addition to some smaller tweaks and stabilization for the models.
A larger change to the data input pipeline was also implemented, which now allows to speed up the data pipeline by moving the batch -> event re-formatting of vector tensors to the worker nodes itself. Todo for the future: this can probably be sped up even further, since the loading of the hdf files could directly provide the data in per-event format.
This PR modifies how SnowStorm values are read in: they are now read in via the SnowstormParametersDict rather than a list of variables that are required to be specified in the correct order. A feature for an arbitrary label key name is also added to the data module, in addition to some smaller tweaks and stabilization for the models.
A larger change to the data input pipeline was also implemented, which now allows to speed up the data pipeline by moving the batch -> event re-formatting of vector tensors to the worker nodes itself. Todo for the future: this can probably be sped up even further, since the loading of the hdf files could directly provide the data in per-event format.