Closed jsadler2 closed 2 years ago
This is the simplest way to implement it, but would we want to try to be more flexible here? Or maybe we just do this for now and then add the flexibility later on down the road?
See my comment above, I think this is still an upgrade from what we have, but ultimately I think we do need more flexibility. I'm fine with incorporating this for now though and continuing to improve on it as we go.
Alright, @SimonTopp - ready for another look. New changes:
nan
in train datanan
in validation dataAlso, for the record, all of the test_data is different because I added two more sites (a total of four now) to make testing these changes more meaningful.
This PR makes the specification of validation and test periods optional. It also allows you to specify validation and test sites in addition to time periods.
Important note and question:
Note: The temporal partitioning happens before the spatial partitioning. Therefore, for the validation and test sites, there is data for only the validation and test time periods, respectively, even though you could argue that data from all time periods would be valid since the model hasn't seen any information from these sites.
Question: This is the simplest way to implement it, but would we want to try to be more flexible here? Or maybe we just do this for now and then add the flexibility later on down the road?
closes #160 closes #161