not-so-random
not-so-random copied to clipboard
Time series neural network performace
Really interested project, I could be interested on using some algorithm like an LSTM to check the performance using other models, have you done that already? If so what were the results?
I will try to make a LSTM version and check the performance over the weekend and keep you posted.
Thanks for the comment! I haven't compared this method with RNNs, but it would be really interesting to check it. I think the toughest obstacle would be implementing learning on the fly. Pre-training on one tester's data likely won't be very effective because of variance in patterns from one person to another. That said I haven't touched RNNs since taking a course a year ago, so maybe I'm mistaken about their computational complexity.