[comp.ai.neural-nets] novice question: selecting training data

jdm5548@tamsun.tamu.edu (James Darrell McCauley) (04/03/91)

Which is better:
  a) selecting training data which is most highly correlated
     to what I am trying to classify, or
  b) randomly selecting training data and averaging the
     performance of the network?

Is there anything wrong with (a)?

I'm using BP with 3-7 variables in the input vector, attempting
to classify data according to an 8th variable.

-- 
jdm5548@tamsun.tamu.edu, jdm5548@tamvenus.bitnet