Intermediate-Deep-Learning-With-Pytorch.md

Deep Learning (24AM654)

Lab Assignment - 2
Course : Intermediate Deep Learning with Pytorch (Datacamp).

Module-1

Training Robust Neural Nets:

1. Pytorch & OOP (10-03-2025, Mon)
2. Optimizers training & Evaluation (11-03-2025, Tue)

---

Pytorhc & OOP: (10-03-2025, Mon)

We use pytorch to define:

• Pytorch Datasets.
• Pytorch Models.

To build a custom dataset:

• Data Loader.

    from torch.utils.data import Dataset  
  
    class WaterDataset(Dataset):
  
        def __init__(self, csv_path):
            
            # below super().__init__() helps make the class behaves like a torch dataset.
            
            super().__init__()
            df = pd.read_csv(csv_path)
            self.data = df.to_numpy()
  
        # we need to create a len function to return the size of the datset.
  
        def __len__(self):
            return self.data.shape[0]
  
        # we need to create a access function "__getitem__" that takes an value as an index (idx) and returns the value at that index (idx), the function returns features and lebels.
  
        def __getitem__(slef, idx):
            features = self.data[idx : -1]
            labels = self.data[idx, -1]
            return features, labels
  

Now we create an instance of a dataset class thats already created and pass the path of the dataset.

    dataset = WaterDatset(
        "__file-path__"
    )

    dataloader_train = DataLoader(
        dataset_train,
        batch_size = 2,
        shuffle = True,
    )

We can use next(iter( )) to iterate over the features & labels.

features, labels =next(iter(dataloader_train))

Now using OOP of python we create a class based model.

    # Net() is the base class for creating a NN in pytorch.
    and we add all the Linear() layers in the __init__ class.

    class Net(nn.module):
        def __init__(self):
            super().__init__()
            self.fc1 = nn.Linear(a, b)
            self.fc2 = nn.Linear(b, c)
            self.fc3 = nn.Linear(c, d)

    # now we create a forward( ) function that makes a forward pass across the network.

    def forward(self, x):
        x = nn.functional.relu(self.fc1(x))
        x = nn.functional.relu(self.fc2(x))
        x = nn.functional.relu(self.fc3(x))

        return x
    net = Net()

---

Optimizers training & Evaluation: (11-03-2025, Tue)