Seq2Seq Attention Models for Timeseries using fastaiv1


I am planning to build a seq2seq network with attention for predicting stock prices using the fastaiv1. Has anyone tried this before?. If yes, could you share some pointers?

I am specifically looking for answers to below questions

  1. Is there an existing DataBunch/ItemList in fastai that can be used for this problem?
  2. All RNN models look tailor-made for NLP in fastai. How to customize to fit them for multi variate timeseries like the stock data.

Any pointers would help.

Thank you


Nisar, i am very involved in this kind of work. Let’s confer. I can be reached at

Hi @nisar009 were you able to find a way to use any of the existing DataBunch/ItemList classes for this?

Nope. I started working on creating Custom ItemList for this. I ll post here the notebook for anyone who needs it.


Hi Nisar this is something I’d be particularly interested in. Would you mind sharing?

I will be interested about such seq2seq model. Is it possible to share your work