Possible to use stack in lazy mode?

[marceloandrioni]

Hello, I was wondering if it is possible to use stack to join two (or more) dimensions, but without loading the data into memory

import xarray as xr

ds = xr.tutorial.open_dataset("air_temperature")
da = ds["air"]
print(da._in_memory)  # False: data is not in memory

da2 = da.stack(latlon=("lat", "lon"))
print(da2._in_memory)

My goal is to use a single index to retrieve a location, instead of using multiple (lat and lon) indexes: da2.isel(station=1000).
When retrieving the data from a TRHEDDS server (like this) it is not possible to load the whole matrix in memory to then subset a single location.

Thank you.

[dcherian]

Not at present I don't think, we only support lazy transpose.

Alternatively, you could use dask/cubed as the underlying array type by passing chunks=. Both of those should support lazy reshape.

[marceloandrioni]

Hello @dcherian , thank you!
Doing ds.chunk().stack(station=["latitude", "longitude"]).isel(station=idx) allowed me to retrieve one or multiple stations without loading the whole array into memory.