AI417UPM/A4_4310813_Ramah

• This repository contains two fully connected neural network models trained on FashionMNIST which are :

  1- dummy.py : This neural network is built from one input layer (784,256), two hidden layers (256,128), (128,100), and one output layer (100,10). The optimizer used was Adagrad with a learning rate of 0.0005.

  2- vanilla.py: This neural network is built from one input layer (784,512), one hidden layer (512,512), and one output layer (512,10). The activation function used between the layers is ReLU(), and the optimizer used is RAdam() with a learning rate of 0.0005.

• To load the models follow the following steps :

  1- insstall dependecies :

  pip install torch huggingface_hub
  

  2-Download and load mode:

    from huggingface_hub import snapshot_download
    import torch as tr
    
    repo = snapshot_download('AI417UPM/A4_4310813_Ramah')
    
    model = tr.hub.load(repo, 'DummyNet', source='local')
    model.eval()
    predictions = model(tr.randn(10000, 784))  
    
    model = tr.hub.load(repo, 'VanillaNet', source='local')
    model.eval()
    predictions = model(tr.randn(10000, 784))