cc3d.statistics fails with signed integer types and floats

[landynfrancis]

If the input to cc3d.statistics is not an unsigned integer, it throws a TypeError: "Function call with ambiguous argument types"

# Works
labeled_data = cc3d.connected_components(np_data)
stats = cc3d.statistics(labeled_data.astype(np.uint16))

# Fails
labeled_data = cc3d.connected_components(np_data)
stats = cc3d.statistics(labeled_data.astype(np.float32))

# Also Fails
labeled_data = cc3d.connected_components(np_data)
stats = cc3d.statistics(labeled_data.astype(np.int16))

This is especially important when using cc3d.dust which returns a numpy array with dtype int16

# Fails, as labeled_data is `int16`
labeled_data = cc3d.dust(np_data)
stats = cc3d.statistics(labeled_data)

# Works
labeled_data = cc3d.dust(np_data)
stats = cc3d.statistics(labeled_data.astype(np.uint16))

[william-silversmith]

cc3d.statistics relies on the input being the output of a cc3d.connected_components call. The input must be numbered from 0 and limited to a reasonable number of elements as the implementation uses direct array indexing for e.g. voxel counting. If your array meets those criteria above, I would recommend using .view or .astype.

I suppose it could be extended, but handling arbitrary labels might be significantly slower since I'd have to use an std::unordered_map or similar. It would require adding a general separate code path for the signed data types. What I could do as a stop-gap is check the bounds on integer types and see if they fall into a reasonable range and then process them as a uint view.

[william-silversmith]

I'm going to release an update that enables compatibility with signed integers so long as the range of the numbers inside is >= 0 and < num voxels.

[william-silversmith]

Ok, this is released!

[landynfrancis]

Great, thank you!