Variable length time series data

nlp
1

Hello all, I got two questions,

  1. I have time series data and each entry has variable time steps. I understand that padding is one way to deal with this. But I wanted to ask if there are better ways to handle this ?
  2. In text data, assume the embedding size is 300. If I’m using LSTMs, Usually, the hidden dimension is smaller than the embedding dimension. However, in my time series data, I have 11 channels (they correspond to embeddings dim in text data). Now is it a good idea to use an RNN with a hidden dim of say 128. I feel like this is not the way to go.
    Any suggestions are welcome.

Thanks in advance.

2

You could organizes your sequences in such a way that all sequences in a batch have the same length using a sampler.