Scott Teresi, 10/27/97, www.teresi.us

Sample runs of my backprop neural net program
on data from the original file "glass.data"


FILES USED:

glass.train	  training set; odd-numbered rows of glass.data file
glass.test	  test set; even-numbered rows of glass.data
glass_norm.train  adjusted input values to all range from 0.0 to 1.0
		  according to the formula x' = (x - xmin)/SD
		  where x is orig. input value, xmin is the smallest it ever
		  gets, SD is the standard deviation of the input set,
		  and x' is the new adjusted value.
glass_norm.test	  adjusted input values to be used for testing

glass.data and glass_norm.data
		  these files contain the train and test data
		  before they were separated


				GLASS DATA

			NORMALIZED	UNCHANGED
						
			Train	Test	Train	Test
Learn	Hidden	Iter-	Set	Set	Set	Set
Rate	Elems.	ations	Accur.	Accur.	Accur.	Accur.

0.05	10-10	10000	69.2	65.4	47.7	46.6

0.05	10	10000	68.2	58.9	50.5	50.5

0.01	5	10000	63.6
0.04	5	10000	61.7
0.05	5	10000	66.4	57.0	45.8	43.9
0.06	5	10000	63.6
0.10	5	10000	48.6
0.20	5	5000	 5.6
0.50	5	5000	 2.8

0.05	2	10000	45.8	45.8	44.9	44.9

0.05	0	10000	44.9	44.9	47.7	48.6

0.05	5	0	21.5