I’ve been playing around with RNNs a lot to do weather predictions. Recently I cam across what you can see below. Training time increasing steadily by 1-2 seconds per epoch. I have run 100’s of experiments with RNNs and this is the first time seeing this.
What are the driving forces behind this?
How can I better set up my experiments so that this does not happen? or does not cause such a big effect?
