Cascade and tensorflow 2

[steto123]

Dear colleagues,

I have not completely given up on the attempt to operate CASCADE as a local version. Unfortunately, TensorFlow 2 is no longer backward compatible. However, a user here has created a modified copy of CASCADE. I have also managed to do the same in terms of syntax, but both variants (predict_nmr_xyzfile.ipynb and the .py version) fail due to the following problem: There appears to be an issue when reading the trained network. The error message is as follows: Cannot assign value to variable 'atomwise_shift/embeddings:0': Shape mismatch. The variable shape (10, 10) and the assigned value shape (10, 1) are incompatible.

I then displayed the structure of the learned network, which was done with:

open HDF5-file

with h5py.File(file_path, 'r') as file:
#show structure
def print_structure(name, obj):
print(f"{name}: {obj}")

#go through all data
file.visititems(print_structure)

and one of the properties is following:
model_weights/atomwise_shift/atomwise_shift_1/embeddings:0: <HDF5 dataset "embeddings:0": shape (10, 1), type "<f4">

But i do not find how can i reshape this in the code.

Has anybody a good idea?

Many thanks

Steffen