Open philosophocat opened 6 years ago
Play with:
In my network with 27 inputs and 1 output I have very fast results with 27 1 1 1 using only 1 training set (row of data) but repeating it quite a lot for very low MSE (> 0.0000001)
Just play with the parameters, after some hours you should get a feeling which settings work best :)
I'm trying to teach network for cubic parabola. It works, but not very good. How can i improve it? Why it works worse, if the configuration is (2, 6, 6, 2) instead of (2, 6, 2), it should be better, shouldn't it?