intro to pytorch

[enryH]

Regarding two questions regarding the exercises. Please see the two comments:

• initialise weights to zero: check out this code -> zero_() method call on paramters (the underscore tell you that is changing the state as far as I know)

• if you make your custom module, it has the attributes you give it (and some methods it inherits). So a nn.Linear instance has attributes weight and bias, not your custom class bases on nn.Module

  class LinearRegression(nn.Module):
  
   def __init__(self):
       """Everything stateful which you need to use your model goes here."""
       super().__init__()  # needs to be here for API reasons -> call nn.Module.__init__
       self.linear_reg = nn.Linear(1, 1)
  
   def forward(self, x):  # now with input
       return self.linear_reg(x)
  
  
   lin_reg = LinearRegression()
   lin_reg.linear_reg.weight # access weight of underlying nn.Linear instance