rule induction

m_r_nour

hi

how neural networks can be used to rule induction?
???

regards
REZA

haddock

m_r_nour

hi

haddock , Here is a forum to ask questions and doubts
so if you do not want help me please don't disturb me  >:(


if we have a decision tree model ,extracting rule induction is easy but how it can be extracted from Neural network? ???

land

Hi!
Haddock, although I always appreciate your comments, I think, this time the manual isn't of any help. And mainly because the problem is simply impossible to solve and nobody thought of writing impossible things into the manual.

So if you could tell me, Reza, how something complex and incomprehensible like neural networks should be transformed to simple rules, I will be happy to implement an operator doing this. Decision trees are convertible only because they are of the same nature as the Rules.
But there's a way of building simply rules form the scratch using the RuleLearner, might be this does suit your needs?

Greetings,
  Sebastian

m_r_nour

Hi Sebastian

thanks for your reply,

but I do not, that is  why I asked this question in data mining part not in problem and support part. I mean my main problem is conceptual problem not software problem.
I do not know is there any way to do that or not? I review many papers and books but non them was useful.


again thanks

REZA

haddock

G'Day Seb,

I think my point was the same as yours. If you read the manual, forum, and tutorial you get a reasonable idea of what can and what cannot be done with unextended RM. There seems to me to be a high correlation between people who ask for stuff that is not in those sources and people who are self-confessed novices. This obliges you, RM staffers, to waste time answering pointless questions, simply because you are too polite and kind. Naturally I am not fettered by such considerations.

On a slightly more pointy note if you Google Scholar on this you'll see that people do try contorted ways to shoe-horn functions into rulesets ( http://www.google.co.uk/search?hl=en-GB&;q=extract+rules+from+neural+networks&sourceid=navclient-ff&rlz=1B3GGGL_en-GBFR315FR315&ie=UTF-8 ). The most convincing apply a NN model to data and then use decision tree builders on the result - what most seem to agree is that you might as well do that on the original data, which of course was your point....