Writing our first training loop

Introduction to Deep Learning with PyTorch

Jasmin Ludolf

Senior Data Science Content Developer, DataCamp

Training a neural network

Create a model
Choose a loss function
Define a dataset
Set an optimizer
Run a training loop:
- Calculate loss (forward pass)
- Compute gradients (backpropagation)
- Updating model parameters

Introducing the Data Science Salary dataset

 experience_level  employment_type  remote_ratio  company_size  salary_in_usd  
        0                0               0.5             1            0.036 
        1                0               1.0             2            0.133     
        2                0               0.0             1            0.234  
        1                0               1.0             0            0.076  
        2                0               1.0             1            0.170

Features: categorical, target: salary (USD)
Final output: linear layer
Loss: regression-specific

Mean Squared Error Loss

MSE loss is the mean of the squared difference between predictions and ground truth

def mean_squared_loss(prediction, target):
  return np.mean((prediction - target)**2)

in PyTorch:

criterion = nn.MSELoss()
# Prediction and target are float tensors
loss = criterion(prediction, target)

Before the training loop

# Create the dataset and the dataloader
dataset = TensorDataset(torch.tensor(features).float(),
                        torch.tensor(target).float())


dataloader = DataLoader(dataset, batch_size=4, shuffle=True)


# Create the model
model = nn.Sequential(nn.Linear(4, 2), 
                      nn.Linear(2, 1))


# Create the loss and optimizer
criterion = nn.MSELoss()
optimizer = optim.SGD(model.parameters(), lr=0.001)

The training loop

for epoch in range(num_epochs):

    for data in dataloader:

        # Set the gradients to zero
        optimizer.zero_grad()

        # Get feature and target from the data loader
        feature, target = data

        # Run a forward pass
        pred = model(feature)
        # Compute loss and gradients
        loss = criterion(pred, target)
        loss.backward()

        # Update the parameters
        optimizer.step()

Let's practice!

Introduction to Deep Learning with PyTorch