I recently modified PredictSunspot.java so that the "sunspots" were just a simple array, counting from 1 to 1000 in integers.
I then trained the network with the first 700, and left 300 to predict.
My results look something like this :
Epoch #357 Error:6.0225646766598225E-6
Epoch #358 Error:5.9548271874202975E-6
Row Actual Predict Closed Loop Predict
701 0.7017 0.6899 0.6899
702 0.7027 0.6906 0.6910
703 0.7037 0.6914 0.6910
My question is - how accurate should the predictions be ? My data set is a very simple linear problem but it fails to spot the pattern that each number increments the previous value by 1.
I've seen NN's do what I consider much more complicated things. Am I doing something wrong, or is this as good as temporal predictions can get ?
Thanks for any replies !