I am quite new to the field.

I am working on a problem involving time-series forecasting of single variable time-series. Data is collected from the pressure sensor on a patient in hospital.

Time window size is fixed to 6 time-points. the target variable is the pressure value at t_7. Rolling window is used

I have 2000 different time-series (each one referred to a different patient). Each time-series has a different length (let's say around 3000 measurements), because the observation time (the time during which the pressure sensor was recording) is different for every patient.

How can I train my LSTM model using all the 2000 time-series?

Thank you!

  • $\begingroup$ How big are the samples? What number of observations are we talking about foe each patient? $\endgroup$ – user2974951 Aug 29 '19 at 6:49
  • $\begingroup$ Thank you for your reply! about 150 observations for each patient $\endgroup$ – andrea_a Aug 29 '19 at 13:11
  • $\begingroup$ Then you can forget about neural networks, you do not have nearly enough data for that. $\endgroup$ – user2974951 Aug 29 '19 at 13:12
  • $\begingroup$ thank you! I was thinking about something similar to their approach .. nature.com/articles/s41586-019-1390-1, there they used recurrent neural networks with multiple time series of different patients, but where each time series, being the time the patient has been in the hospital, not very long. thay also grouped the parameters into six hours blocks, so it is like having for every 24 hours, 6 measurements. Imagine a 30 days stay in the hospital (already long), it is 180 measurements. this is why I thought using this approach to my problem $\endgroup$ – andrea_a Aug 29 '19 at 13:14
  • $\begingroup$ Unfortunately that article is behind a paywall so we cannot access it. Maybe you can try summarizing it in your question, how they achieved what they did, and how you want to replicate it. $\endgroup$ – user2974951 Aug 29 '19 at 13:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.