Saving numpy arrays with HDF5

I/O
1

I have just switched from fenics 2017 to 2018.1 and am getting an error when I try to save an array using HDF5:

self.myfile.write(np.array([float(self.checkpointInt)]), ‘checkpointInt’)
AttributeError: ‘numpy.ndarray’ object has no attribute ‘_cpp_object’

It seems that fenics no longer supports the saving of np arrays or ints or floats in HDF5. I just need to save an int using the HDF5 format in fenics. Anyone have any ideas how to do this with the 2018 version?

Thanks,
Jared

2

I’m also running into this problem currently. Were you ever able to resolve this?

3

This probably wasn’t answered because there was no attempt at providing a MWE or context.

You could probably just use h5py.

4

It sounds like a library version mismatch. Perhaps your new dolfin wasn’t compiled for the numpy that you’ve got installed. Mismatches like that can happen if pip is used to install packages.

5 
AttributeError: ‘numpy.ndarray’ object has no attribute ‘_cpp_object’

indicates attempting to pass a numpy array to a dolfin function which requires a Function or some other backend type (i.e. which has a _cpp_object). I doubt it’s a versioning issue.

The issue is we don’t know what self.myfile is. I’m assuming dolfin.HDF5File and not a h5py.File.

6

True. Not enough information to be certain.

1 Like