Function for spatiotemporal train/validation/test split from a large xarray dataset

[CoenvdE]

Hi everyone,

I’d like to start a discussion about adding functionality to create static spatiotemporal train, validation, and test splits from a single large xarray dataset (e.g., global-scale data).

The goal is to generate these splits based on:
• The spatial and temporal size of each sample
• The stride between samples
• A list of user-defined validation and test regions (as static spatiotemporal holdouts)

The expected output would be, for each split, a list of dictionaries containing the coordinates (start and end) of each slice. This structure would make it straightforward to iterate over samples in a dataloader.

Keeping the validation and test regions static is important to ensure consistent model evaluation and comparability across experiments.

I might already be working on a prototype for this, but I wanted to open the discussion first to gather feedback on the design and identify possible integration points.