I am trying to build a RNN that can take time series as input for example from various sensors. While building the RNN there are a couple of initial lengths that must be defined. The size of the cell and the length of how much you will unroll the RNN.

From what i understand, cell size is arbitrary and can be modified independent of how the inputs are shaped. But how much you will unroll the RNN depends on your input length.

If i have a time series of 1 hour consisting of 1000 values, should i choose to unroll the RNN 1000 times, or is it too much? My goal is to classify the whole time series, and ideally i want all 1000 time steps to be taken into account for this classification.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.