From: Henry W
Subject: Info on decision-trees
Date: 
Message-ID: <37643c3d.15263847@nntpserver.swip.net>
Hello!
I wonder if anyone can advise me on where to find information on how
to handle missing attribute values in decision-tree learning
algorithms (Specifically how to implement it in Common Lisp). 
Either on the Internet, or in an orderable book.
Thanks in advance!

/Henry
········@hotmail.com

From: Dennis Dunn
Subject: Re: Info on decision-trees
Date: 
Message-ID: <376472DD.A4D5C229@cwix.com>
Henry W wrote:
> 
> Hello!
> I wonder if anyone can advise me on where to find information on how
> to handle missing attribute values in decision-tree learning

Chapter 18, "Artificial Intelligence" by Russell & Norvig talks about
inductive learning.  Exercise 18.11 has you modifying one of the
algorithms in the text to account for missing attributes.

I have also read that you can assign the value which has the highest
frequency for that attribute.  I wish I could remember where I read
that, I've been through so many lately they all seem to run together.

Good luck.

--dennis
From: Henry W
Subject: Re: Info on decision-trees
Date: 
Message-ID: <3764e559.3148885@nntpserver.swip.net>
>> I wonder if anyone can advise me on where to find information on how
>> to handle missing attribute values in decision-tree learning
>
>Chapter 18, "Artificial Intelligence" by Russell & Norvig talks about
>inductive learning.  Exercise 18.11 has you modifying one of the
>algorithms in the text to account for missing attributes.

>I have also read that you can assign the value which has the highest
>frequency for that attribute.  I wish I could remember where I read
>that, I've been through so many lately they all seem to run together.

I have read and reread chapter 18 in that book, but it doesn't make
things much clearer, though 18.11 is pretty much what I'm trying to
do. I'm having problems with the whole concept of "weighting" the
attributes.

Just using the value with the highest frequency would undoubtedly be
easier to implement, but wouldn't the result (the tree's output) be
misleading?

/Henry
·········@hotmail.com