How to get a intermediate value from nn.Sequential?

vision
1

Fox example,if I have a layer:

nn.Sequential(
nn.Linear(1,1),
nn.ReLU(),
nn.Linear(1,1),
nn.ReLU(),
nn.Linear(1,1),
)

My question is how can i got a output form the second nn.Linear(1,1) from this Sequential

2

oh I got it.I can use .children()

3 Likes
3

@mshmoon, how do you do that more specifically? I’m dealing with the same problem, so your advice’d be appreciated. Thanks a lot in advance!

1 Like
4

What kind of issue are you facing?

You could do something like below:

seq = nn.Sequential(nn.Linear(5,5), nn.Dropout(0.2), nn.Linear(5,1))
x = torch.rand(5,5)
for layer in seq:
     x = layer(x)
     print(x)
5 Likes
5

Hi @InnovArul,
The problem I was concerned with was getting output of the hidden layers of AlexNet. It turned out that @fmassa has already provided a simple solution:

2 Likes
7

Thanks your question & solution page helped me to find solution

And the code below works for me

> fnn.model

Sequential(
  (0): Linear(in_features=4, out_features=3, bias=True)
  (1): ReLU()
  (2): Linear(in_features=3, out_features=3, bias=True)
  (3): LayerNorm()
  (4): ReLU()
  (5): Softmax()
)


> fnn.model[:-1](data_torch_variable))

i.e. 
fnn.model[:-1](pre.trans_to_variable(data_x))[0:3]

tensor([[0.0000, 0.5921, 0.4040],
        [0.0000, 0.5703, 0.4293],
        [0.0000, 0.5618, 0.4390]], grad_fn=<SliceBackward>)
3 Likes
8

How does .children() work?

2 Likes