When I did a for loop + .load() on my scipy presentation I noticed that there are discernible gaps in the dask stream which indicate to me that the graph setup is happening for each datasets separately.
I wonder if datatree could just 'append' each computation to a graph and then execute that large graph to be more efficient?
When I did a for loop +
.load()
on my scipy presentation I noticed that there are discernible gaps in the dask stream which indicate to me that the graph setup is happening for each datasets separately.I wonder if datatree could just 'append' each computation to a graph and then execute that large graph to be more efficient?
Related to https://github.com/xarray-contrib/datatree/issues/97