Closed corranwebster closed 10 years ago
int32 should be more than sufficient.
It looks like the types vary depending on the field and it wouldn't be too much work to go through the spec and set the corresponding type per field.
Is there any reason not to use most specific type possible?
It would reduce the memory footprint of the metadata somewhat to do that but since the the bulk of the space is taken by the image arrays I doubt it would be significant. Numpy should upcast stuff appropriately as needed if we do any calculations on the metadata. I don't have a strong opinion eitehr way.
Currently integer fields are specified as being type
np.int
, but this varies depending on whether the platform is 64 or 32 bit. It's probablynp.int32
that you want.There's a similar issue with floats, as well.