Adding a pro tip about the dangers of using geographical chips created from overlapping strides and then randomly split into train/validation/test sets. Need to raise more awareness that the 'geo' in GeoML matters, because spatial statistics is special 😝 Also fixed an incorrect link to torchdata.datapipes.iter.Batcher.
Adding a pro tip about the dangers of using geographical chips created from overlapping strides and then randomly split into train/validation/test sets. Need to raise more awareness that the 'geo' in GeoML matters, because spatial statistics is special 😝 Also fixed an incorrect link to torchdata.datapipes.iter.Batcher.
Preview at https://zen3geo--43.org.readthedocs.build/en/43/chipping.html#slicing-with-xbatcherslicer
Xref https://github.com/pangeo-data/xbatcher/discussions/78#discussioncomment-3387295