Hello,
Does anyone know where I can find a Sequence to Sequence example with multiple features, in which the input sequence has a different number of timesteps than the output sequence? For example feed 5 days of data, to forecast 2 days.
Other than using LastTimeStep() I haven’t been able to figure out how to input one shape and have the output as a different shape.
So given the shape is: [batchSize, features, timesteps]
I’m hoping to do [10, 4, 5] → [10, 1, 2] for example.
4 features and 5 timesteps, to forecast 1 feature with 2 timesteps.
Is this possible? Or do the input/output have to match? Or I can pad output with 0’s where I don’t want predictions?
Thanks!