How to model time-series data using recurrent neural network or any other model?

Question

I have a N data samples where each sample is a signal of varying length. The important information relating to associated classlabel could be anywhere in that signal? As such, I don't want to chop of some of my signals to make all of them equal length. People suggested me to use recurrent neural network but I can't formalize my problem with RNN. In NLP (where I see most of RNNs being used) they want to predict a word looking at previous words but here my whole signal has one label not different labels.

Looking forward for some insightful answers.

Thanks.

Answer 1

Have a look at this tutorial: http://machinelearningmastery.com/sequence-classification-lstm-recurrent-neural-networks-python-keras/

This is NLP but the goal is to classify text instead of predicting the next word. As a matter fact it is much easier to do the classification since you can just feed the output of RNN layer into softmax layer with 2 outputs (class 0 or 1)