readme.txt

Revision 27, 356 bytes (checked in by qbert, 5 years ago)
Working 1.1, no serialize.d though

Line
1	Common hidden layer neuron count is in the range ( -A / sqrt(N) , A / sqrt(N) ) , where N is number of inputs, and A is a constant 2 or 3. This helps elimiinate saturation.
2
3	Typical learning rate is between 0.05 <= n <= 0.75 ( 0.1 is common , its also common to scale the learning rate relative to the training set )
4	Typical Momentum is 0 <= a <= 0.9

Note: See TracBrowser for help on using the browser.

and