RNNs Masking Layer - does the "-1" order matters?

TensorFlow General Discussion
1

Hi. I’m currently developing a GRU with a masking layer “-1”.
My sequence shapes are (5, 2048) which means sequence length 5 of 2048 features. Imagine that my sequence is something like: baseline, month6, month12, month18 and month24.
Now from that sequence i want to create new ones: baseline, month6, month12, month18. Where should I pad with “-1”? Option A or Option B?
Option A: baseline, month6, month12, month18, -1
Option B: -1, baseline, month6, month12, month18

Is this a good approach for generating subsequences?

3

Hi @Goncalo_Galante ,

Padding at the end of the sequence is generally the preferred approach.This is the standard approach in most deep learning frameworks and makes it easier for the model to learn meaningful patterns.

Thanks.