Great post! (As usual). You pointed a several changes to improve the results. I would also add:
I wonder how significant is the improvement in your ‘validation’ score. You’re comparing point estimates. But what happens if you try quantify uncertainty around your validation scores using resampling techniques for instance? Especially because the sample size is small. Some ideas here: http://www.fharrell.com/post/split-val/
Thank you so much for this well-documented article, I find it useful and easy to follow. I understand it is just the starting point to predict time series by comparing LSTM and ANN performances. I think the next step would be to increase the input size of the LSTM and predict the next value from a wider window over the past.
Hi Susan, nice article! I’m working on something similar so I really liked reading what you did :).
If I may offer one suggestion: I see that to calculate the pairwise similarity matrix you make a new empty array, and then loop over each pair of elements to calculate the pairwise scores. This makes sense, however the…