FCHL19 padding

[jvschleicher]

In the previous version (qml) this function was used to generate fchl19 representations:
reps = np.array([qml.representations.generate_fchl_acsf(x.nuclear_charges, x.coordinates, pad=max_natoms,
elements=atomtypes) for x in mols])
Here, the padding is nonexistent, which leads to shape issues in the rep generation:
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions.

[kvkarandashev]

Thank you for the feedback! We have been removing need for padding in everything QML2-related to make things less memory-inefficient, hence absence of the option. So, please consider using:

reps = [qml2.representations.generate_fchl(x.nuclear_charges, x.coordinates) for x in mols]

Generating local kernels from such lists can be done analogously to QML1 using this function:
https://github.com/qml2code/qml2-dev/blob/54625fe7e14b1dfb0606882b829202cf2daa113b/examples/old_qml_interfaces/old_qml_interfaces.py#L25
(We didn't include it in the main code because we wanted to mainly restrict ourselves to functions that operate on NumPy arrays, but that might change in the future.)

Also note how merging arrays of inhomogeneous shape is avoided in this example of using structured orthogonal random features with FCHL19:
https://github.com/qml2code/qml2-dev/blob/main/examples/sorf/ex_sorf.py