Graph construction with Bool features

I am constructing graph using gconstruct and I download my data which contains boolean features. There features are the result of one-hot-encoding and they truly can be either 0 or 1.

During the call of gconstruct, I get an exception, that boolean type is not supported (graphstorm/gconstruct/transform.py", line 664):

       assert np.issubdtype(feats.dtype, np.integer) \
                or np.issubdtype(feats.dtype, np.floating), \
                f"The feature {self.feat_name} has to be integers or floats."

I can easily fix that problem by adding an extra check for a value to be of boolean type:

       assert np.issubdtype(feats.dtype, np.integer) \
                or np.issubdtype(feats.dtype, np.floating) \
                or np.issubdtype(feats.dtype, np.bool_), \
                f"The feature {self.feat_name} has to be integers or floats or bools."

The question if this is a valid thing to do? I don't want to recompress data back to int32 because for one-hot-encoding of multiple categories that is wasteful in terms of disk space and networking, but I am not 100% sure that graphstorm does transformation right after adding this condition.

Can someone check if this fix is correct and maybe merge into the main branch?

Thank you.

Start in graphstorm/gconstruct/transform.py around line 664 and trace how feature arrays are handled after the dtype validation. Run gconstruct with boolean one-hot features and compare the resulting transformed data with the integer and floating-point paths. Done means valid boolean features are accepted and transformed correctly without weakening validation for unsupported types.