A quick workflow demo in PyTorch implementing linear regression

Model

Use nn.Parameter

class LinearRegressionModel(nn.Module):
  def __init__(self):
     super().__init__()
     self.weights = nn.Parameter(torch.randn(1, requires_grad=True, dtype=torch.float))
     self.bias = nn.Parameter(torch.randn(1, requires_grad=True, dtype = torch.float))

  def forward(self, x: torch.Tensor) -> torch.Tensor:
      return self.weights * x + self.bias

model =  LinearRegressionModel()

Use nn.Linear

class LinearRegression(nn.Module):
  def __init__(self):
    super().__init__()
    self.linear_layer = nn.Linear(in_features = 1, out_features=1)

  def forward(self,x: torch.Tensor)-> torch.Tensor:
    return self.linear_layer(x)

model =  LinearRegressionModel()

Loss function

loss_fn = torch.nn.L1Loss()
optimizer = torch.optim.SGD(params=model_0.parameters(), lr= 0.01)

Training loop

epochs = 300

for epoch in range(epochs):
  model.train() #set up the mode. Train mode in Pytorch sets all parameter to requite grad that require the gradients 
 
  y_pred = model(X_train)  #forward pass

  loss = loss_fn(y_pred, y_train) # calculate loss

  optimizer.zero_grad()# Assign 0 grad so gradients don't get accumulated at each step 
 
  loss.backward() # perform back propagation

 
  optimizer.step()  # Update the parameters according to the gradients.

  
model.eval()  #turn off settings not needed for evaluation
  
with torch.inference_mode():
    test_pred = model(X_test)

    test_loss = loss_fn(test_pred,y_test)
badge