Hi,
I'm using Smilearn to learn the parameters of a naive bayes classifier. There are no missing values in the data and one of the attributed has only one possible value.
However, when I learn the parameters and I look at the .xdsl file generated for the variable that has only one value the file shows a probability distribution for that variable taking two values: one value the true value it takes and the other one "state_unobserved". I copy and paste the relevant part of the file here:
<cpt id="D">
<state id="State_5" />
<state id="State__unobserved" />
<parents>G</parents>
<probabilities>0.9841269841269841 0.01587301587301587 0.9857142857142857 0.01428571428571429 0.9814814814814814 0.01851851851851852</probabilities>
</cpt>
Why is this happening? Can Smilearn not deal with variables that take only one value?
Thank you very much,
Noelia
Unobserved state?
-
- Site Admin
- Posts: 1417
- Joined: Mon Nov 26, 2007 5:51 pm
Re: Unobserved state?
The 'unobserved state' is the way for SMILEarn to deal with data column containing only one value. SMILE's discrete nodes can have two or more outcomes and SMILearn must follow that constraint.ermutarra wrote:Why is this happening? Can Smilearn not deal with variables that take only one value?