implementing _Training_Neural_Network_with_PyTorch .pdf

implementing _Training_Neural_Network_with_PyTorch .pdf

• 1.
  Implementing and Traininga Neural Network with PyTorch Building and Training a Feed-Forward Neural Network on the MNIST Dataset By KAHSAY August 28, 2024 1
• 2.
  Introduction
• 3.
  Overview • In thispresentation, we will go through the process of building and training a feed-forward neural network using PyTorch. • Our objective is to classify handwritten digits using the MNIST dataset. • We’ll define each step, method, and function in the implementation. 2
• 4.
  Step 1: Downloadingthe Dataset
• 5.
  Downloading the Dataset •The download_mnist_datasets function is responsible for downloading the MNIST dataset. • The MNIST dataset contains images of handwritten digits (0-9) and is commonly used for training image processing systems. 1 def download_mnist_datasets(): 2 train_data = datasets.MNIST( 3 root="data", 4 train=True, 5 download=True, 6 transform=ToTensor(), 7 ) 8 validation_data = datasets.MNIST( 9 root="data", 10 train=False, 11 download=True, 12 transform=ToTensor(), 3
• 6.
  Step 2: Creatingthe Data Loader
• 7.
  Creating the DataLoader • The create_data_loader function creates a data loader for the training data. • Data loaders are used to batch and shuffle the data, making it easier to manage during training. 1 def create_data_loader(train_data, batch_size): 2 train_dataloader = DataLoader(train_data, batch_size =batch_size) 3 return train_dataloader 4 4
• 8.
  Step 3: Buildingthe Model
• 9.
  Building the Model •The FeedForwardNet class defines the architecture of our feed-forward neural network. • The network consists of a flattening layer, two dense (fully connected) layers, and a softmax layer for output. 1 class FeedForwardNet(nn.Module): 2 3 def __init__(self): 4 super().__init__() 5 self.flatten = nn.Flatten() 6 self.dense_layers = nn.Sequential( 7 nn.Linear(28 * 28, 256), 8 nn.ReLU(), 9 nn.Linear(256, 10) 10 ) 11 self.softmax = nn.Softmax(dim=1) 12 13 def forward(self, input_data): 5
• 10.
  Step 4: Trainingthe Model
• 11.
  Training the Model •The train_single_epoch method trains the model for a single epoch. • It performs a forward pass, computes the loss, backpropagates the error, and updates the model weights. 1 def train_single_epoch(model, data_loader, loss_fn, optimiser, device): 2 for input, target in data_loader: 3 input, target = input.to(device), target.to( device) 4 5 # calculate loss 6 prediction = model(input) 7 loss = loss_fn(prediction, target) 8 9 # backpropagate error and update weights 10 optimiser.zero_grad() 11 loss.backward() 6
• 12.
  Training the Model(Continued) • The train function controls the training process over multiple epochs. • It repeatedly calls the train_single_epoch function and prints the training progress. 1 def train(model, data_loader, loss_fn, optimiser, device , epochs): 2 for i in range(epochs): 3 print(f"Epoch {i+1}") 4 train_single_epoch(model, data_loader, loss_fn, optimiser, device) 5 print("---------------------------") 6 print("Finished training") 7 7
• 13.
  Step 5: Savingthe Model
• 14.
  Saving the Model •After training, we save the trained model’s parameters using torch.save. • The model can be loaded later for inference or further training. 1 torch.save(feed_forward_net.state_dict(), " feedforwardnet.pth") 2 print("Trained feed forward net saved at feedforwardnet. pth") 3 8
• 15.
  Conclusion
• 16.
  Conclusion • In thispresentation, we covered how to implement and train a neural network using PyTorch. • We used the MNIST dataset for training a simple feed-forward neural network. • Each method and function was explained in detail, showing how they contribute to the overall process. 9