I have a time series with both continuous and categorical features, and I want to do a prediction task.

I will elaborate:

The data is composed of 100Hz sampling of some voltages, kind of like an ecg signal, and of some categorical features such as "green", "na" and so on.

In total, the number of features can reach 300, of which most are continuous.

The prediction should take in a chunk of frames and predict a categorical variable for this chunk of frames.

I want to create a deep learning model that can handle both categorical and continuous features.

Best I can think of is two separate losses, like MSE and cross entropy, and a hyperparameter to tune between them, kind of like regularization.

Best I could find on this subject was this, with an answer from 2015.

I wonder if something better was invented, since then, or maybe just someone here knows something better.

  • $\begingroup$ Can you please provide more info about your actual data? $\endgroup$ – nbro Jan 5 at 0:36
  • $\begingroup$ @nbro sure, like what? I can describe everything, but not sure what exactly is needed, please say and I will add. $\endgroup$ – Gulzar Jan 5 at 0:40
  • 1
    $\begingroup$ Ok, now, after thinking about it a little more, I realize that you want to predict what's going to come next in the time-series data, and that's why you thought about using two loss functions. That wasn't initially clear to me, and that's why I asked you to provide more details about the data, but I guess that's no longer strictly needed. In any case, it may be useful to know the nature of your data, e.g. if they are sequences of images and numbers or whatever. $\endgroup$ – nbro Jan 5 at 0:45
  • $\begingroup$ @nbro I now see I was not clear. I edited, hope it is better now. $\endgroup$ – Gulzar Jan 5 at 10:47

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.