why do u have <div class="snippet-clipboard-content notranslate position-relative

is: <div class="snippet-clipboard-content notranslate position-relative overflow-a

Here is the net without setattr() <div class="sni

Why do u have 'setattr(self, 'fc%i' % i, fc)' about pytorch-tutorial HOT 2 CLOSED

morvanzhou commented on July 28, 2024

Why do u have 'setattr(self, 'fc%i' % i, fc)'

from pytorch-tutorial.

Comments (2)

brando90 commented on July 28, 2024

is:

self._set_init(fc)

needed?

from pytorch-tutorial.

MorvanZhou commented on July 28, 2024

Here is the net without setattr()

Net(
  (bn_input): BatchNorm1d(1, eps=1e-05, momentum=0.5, affine=True)
  (bn0): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (bn1): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (bn2): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (bn3): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (bn4): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (bn5): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (bn6): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (bn7): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (predict): Linear(in_features=10, out_features=1)
)

Here is the net with that:

Net(
  (bn_input): BatchNorm1d(1, eps=1e-05, momentum=0.5, affine=True)
  (fc0): Linear(in_features=1, out_features=10)
  (bn0): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (fc1): Linear(in_features=10, out_features=10)
  (bn1): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (fc2): Linear(in_features=10, out_features=10)
  (bn2): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (fc3): Linear(in_features=10, out_features=10)
  (bn3): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (fc4): Linear(in_features=10, out_features=10)
  (bn4): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (fc5): Linear(in_features=10, out_features=10)
  (bn5): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (fc6): Linear(in_features=10, out_features=10)
  (bn6): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (fc7): Linear(in_features=10, out_features=10)
  (bn7): BatchNorm1d(10, eps=1e-05, momentum=0.5, affine=True)
  (predict): Linear(in_features=10, out_features=1)
)

You now see the difference.
For your second question, my answer is we need _set_init() in this tutorial to witness how bad initialization could affect the result. But in your own application, you don't need it.

from pytorch-tutorial.

Recommend Projects