An example of where it would be useful to be able to save several dask arrays together:
You can still return the header by just setting the key="header" for a second memmap_distributed call. It will add some time onto the saving of the dataset as the entire dataset might get loaded into ram with most of it thrown away.
Only once. That will merge taskgraphs as necessary and might reduce the time for saving certain signals. I've thought about it for things like saving lazy markers of possibly creating a hs.save() function for handling mulitple signals if you wanted to save multiple parts of some anaylsis efficently. This is a fairly abstract/higher level concept so maybe it would be seledomly used.
An example of where it would be useful to be able to save several dask arrays together:
You can still return the header by just setting the
key="header"
for a secondmemmap_distributed
call. It will add some time onto the saving of the dataset as the entire dataset might get loaded into ram with most of it thrown away.Really what we should do is add things to a
to_store
context manager and then call: https://github.com/hyperspy/rosettasciio/blob/31bd677cc4c02c5787ae9b61250bc1431f352cba/rsciio/hspy/_api.py#L111Only once. That will merge taskgraphs as necessary and might reduce the time for saving certain signals. I've thought about it for things like saving lazy markers of possibly creating a
hs.save()
function for handling mulitple signals if you wanted to save multiple parts of some anaylsis efficently. This is a fairly abstract/higher level concept so maybe it would be seledomly used._Originally posted by @CSSFrancis in https://github.com/hyperspy/rosettasciio/pull/267#discussion_r1624876062_