Define datatype for make_geocube

[mjyshin]

Hello, I have a lot of Shapely polygons as shapes that I want to rasterize into an output image, but the output will be huge. Is there a way to use make_geocube so that when I run something like

gdf = gpd.GeoDataFrame({'mask': shape_types}, geometry=shapes)
cube = make_geocube(gdf, resolution=(-1, 1), fill=0)
mask = cube.mask.values

the mask will be an int8 array instead of float64 (shape_types will naturally only be small integers)? Thanks in advance!

[snowman2]

You can convert it afterwards:

mask = cube.mask.values.astype("int8")

[mjyshin]

Thank you for the reply! I was hoping to bypass some memory issues; is there no way to specify it beforehand to ensure Numpy won't try to initialize a giant ndarray of floats?

[snowman2]

It is currently defaulted to float64 ref. If you would like to specify a different dtype, you could open up a PR to add the dtype kwarg to rasterize_image and then use functools.partial to change the dtype similar to what is done in this example ref.

[snowman2]

See: #82 and #98