Model code issues

[LanX1ang]

Why is it that in Models.py, the direct FedAvg approach is still used during the federated learning global model update, instead of combining the parameters clustered by cosine similarity and affinity propagation