Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Causality

See original GitHub issue

Hi, In which part of the code you make sure the network is causal? Is it the Chomp1d ?

Thanks a lot! Amir

Issue Analytics

State:
Created 5 years ago
Reactions:1
Comments:8 (3 by maintainers)

Top GitHub Comments

6reactions

jerrybai1995commented, Apr 25, 2018

Please see Figure 1(a) in our paper for reference. We pad by (k-1)*d on the two sides of the input for convolution, and then use Chomp1d to remove the (k-1)*d output elements on the right.

If you draw a figure yourself, you would find that this is equivalent to removing the “future elements”, which ensures causality.

4reactions

Songweipingcommented, Apr 25, 2018

Thanks for the prompt reply.

Do you mean you pad by (k-1)*d zeros on both sides of input? It seems only on left side in Figure 1(a). Besides, padding on left side makes sense to me. I must misunderstand something.

Will there be some leakage if we only pad on left side without Chomp1d? Based on my current understanding, if we only pad (k-1)*d elements on left side, convolutional layer will produce a sequence with the same length of input. Is there something wrong?

Thanks!