algorithms/IRM.py

• 8 E501: line too long (100 > 79 characters) (and 3 similar)

  • Line too long (100 > 79 characters):

    18 author={Arjovsky, Martin and Bottou, L{\'e}on and Gulrajani, Ishaan and Lopez-Paz, David},

  • Line too long (97 > 79 characters):

    56 self.config = config # Need to store config for IRM because we need to re-init optimizer

  • Line too long (97 > 79 characters):

    58 assert isinstance(self.loss, ElementwiseMetric) or isinstance(self.loss, MultiTaskMetric)

  • Line too long (87 > 79 characters):

    61 grad_1 = autograd.grad(losses[0::2].mean(), [self.scale], create_graph=True)[0]

  • Line too long (87 > 79 characters):

    62 grad_2 = autograd.grad(losses[1::2].mean(), [self.scale], create_graph=True)[0]

  • Line too long (91 > 79 characters):

    69 # this returns the average loss and penalty across groups, regardless of group size

  • Line too long (91 > 79 characters):

    70 # But the GroupLoader ensures that each group is of the same size in each minibatch

  • Line too long (91 > 79 characters):

    76 for i_group in group_indices: # Each element of group_indices is a list of indices