Forecasting a time series about neuralcde HOT 4 CLOSED

I think you're going about this in the wrong way. Neural CDEs are essentially "continuous time RNNs", so have a think about how you'd solve the problem using RNNs first -- then switch them out for differential equations.

In this case, CDEs take a path as an input. If your path isn't defined in the region [2, 4], then how can it take it as an input?

What you probably want is a seq2seq model. Run a CDE over the region [0, 2]. Use the result of that as an initial condition for an ODE over the region [2, 4].

from neuralcde.

I think I miss phrases my goal, let me restate it:
X = my input which is within time [0,2] sec
y = my target which is withing time[2,4]

Can I integrate over cdeint t=[2,4] even if the input is only within [0,2]?

from neuralcde.

And what I'm saying is that the answer is no. You should integrate over [0,2] only.

from neuralcde.

Thanks, makes sense what u said earlier about seq2seq!

from neuralcde.