I want to predict two separate y-values (not really logically connected) based on an input sequence of data (values x). Using LSTM cells.

Should I train two models separately or should I just increase the dimension of the last layer to 2 instead of 1 and feed the fitting algorithm y-values as 2D pairs? In other words, will there be a lot of interference or can I expect on average similar results with one or two models?

  • 1
    $\begingroup$ Just to be clear, there is only one x sequence for each two different y values? The two y values are therefore two different kinds of thing associated with the same type of x sequence? $\endgroup$ – Neil Slater Feb 25 at 20:51
  • $\begingroup$ yes. For example, if I am trying to predict the next value of the sequence m step ahead and n steps ahead, n > m. Another example is I am observing a multi-valued function $f:R \rightarrow R^2$ and the values are somewhat related and I'm trying to predict a new pair. $\endgroup$ – Jake B. Feb 26 at 16:52

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.