Running a forward pass

Introduction to Deep Learning with PyTorch

Jasmin Ludolf

Senior Data Science Content Developer, DataCamp

What is a forward pass?

Input data flows through layers
Calculations performed at each layer
Final layer generates outputs

Outputs produced based on weights and biases
Used for training and making predictions

Forward pass representation

What is a forward pass?

Possible outputs:

Binary classification
Multi-class classification
Regressions

Forward pass representation with final output highlighted

Binary classification: forward pass

Coding block with added comments

# Create binary classification model
model = nn.Sequential(
  nn.Linear(6, 4), # First linear layer
  nn.Linear(4, 1), # Second linear layer
  nn.Sigmoid() # Sigmoid activation function
)

Binary classification: forward pass

# Pass input data through model
output = model(input_data)
print(output)

tensor([[0.5188], [0.3761], [0.5015], [0.3718], [0.4663]], 
    grad_fn=<SigmoidBackward0>)

Output: five probabilities between 0 and 1, one for each animal
Classification (0.5 threshold):
- Class = 1 (mammal) for values ≥ 0.5 (0.5188, 0.5015)
- Class = 0 (not mammal) for values < 0.5 (0.3761, 0.3718, 0.4633)

Multi-class classification: forward pass

Class 1 - mammal, class 2 - bird, class 3 - reptile

n_classes = 3


# Create multi-class classification model
model = nn.Sequential(
  nn.Linear(6, 4), # First linear layer
  nn.Linear(4, n_classes), # Second linear layer

  nn.Softmax(dim=-1) # Softmax activation 
)


# Pass input data through model
output = model(input_data)
print(output.shape)

torch.Size([5, 3])

Multi-class classification: forward pass

Each row sums to one
Predicted label = class with the highest probability
Row 1 = class 1 (mammal), row 2 = class 1 (mammal), row 3 = class 3 (reptile)

Regression: forward pass

# Create regression model
model = nn.Sequential(
  nn.Linear(6, 4), # First linear layer
  nn.Linear(4, 1) # Second linear layer
)

# Pass input data through model
output = model(input_data)

# Return output
print(output)

tensor([[0.3818],
        [0.0712],
        [0.3376],
        [0.0231],
        [0.0757]], 
        grad_fn=<AddmmBackward0>)

Let's practice!

Introduction to Deep Learning with PyTorch