last commit error

[thegodone]

training % python3 train_graph.py --hyper hyper/hyper_esol.py --category GIN

Traceback (most recent call last):
File "/Users/tgg/Github/gcnn_keras4/training/train_graph.py", line 9, in
import kgcnn.losses.losses
ModuleNotFoundError: No module named 'kgcnn.losses'

I've updated github last code and only comment "pyxtal>=0.5.5" installation in the setup.py file (it needs pyshtools installation not compatible with Apple Silicon system).

[PatReis]

Hello, thanks.
I am sorry, but I do not understand the error. The module should be there: https://github.com/aimat-lab/gcnn_keras/tree/master/kgcnn/losses
Is it really the latest git version? I am not yet finished with version 4.0. So it is not available on pip yet.
Can you check the installation agian?

[thegodone]

I found the issue basically the version 4.0 is linked to keras 3 and keras 3 is not supported by Apple silicon chips. What is the last commit before the Keras 3 jump ? Le ven. 8 déc. 2023 à 11:55, Patrick Reiser ***@***.***> a écrit :

…

Hello, thanks. I am sorry, but I do not understand the error. The module should be there: https://github.com/aimat-lab/gcnn_keras/tree/master/kgcnn/losses Is it really the latest git version? I am not yet finished with version 4.0. So it is not available on pip yet. Can you check the installation agian? — Reply to this email directly, view it on GitHub <#120 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJBWYWK4Z2ZUR6FSNSWSC3YILW33AVCNFSM6AAAAABAMMCGMSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNBWHE2TSMBUGM> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[PatReis]

The last version would be 3.1.0 but I do not think that current trainng scripts will work with 3.1.0.
The step to keras 3 is actually quite substantial. Do you know what exactly does not work with keras 3?
Is it the tensorflow version ?

[PatReis]

If you want I can update your scripts to kgcnn 4 and keras 3.

[PatReis]

I think keras 3.0 should be fine for all OS. You can install an older tensorflow version like e.g. 2.14 which installs keras 2.
But then you can uninstall keras and install keras 3 without chaning tensorflow. Keras 3.0 worked also with tensorflow 2.14 as far as I can tell.

[thegodone]

I changed two things basically and now it works for tf 2.15 / v3.1.0

# from kgcnn.utils.models   
from kgcnn.model.utils import get_model_class # kgcnn.utils.models decrepeated starting 2.2.4
from kgcnn.molecule.encoder import OneHotEncoder # remove starting 3.0.1
from kgcnn.training.hyper import HyperParameter # moved from hyper to training