IBk and IB1

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

IBk and IB1

Santi Planet
Mensaje
Hello,
 
I don't understand very well the differences between IBk and IB1. I have found this in the wekalist:
 
"There are a couple of differences. The biggest one is probably that IBk
will keep extending the list of neighbours when several instances are
equally far away. This happens quite frequently on datasets with
nominal attributes (e.g. more than one neighbour might be used for
prediction if k=1). IB1 doesn't do this and uses the first nearest
neighbour it finds."
 
Thinking about nominal attributes, I understant that, in IBk, you choose "several instances if they are equally far away", and then you choose the majority class of these instances, while, in the same case, IB1 only chooses "the first nearest neighbour it finds"... but if you had three instances equally far away, two of them belonging to class A and the other belonging to class B, which instance would you choose? All of the three instances are equally far away so I suppose that IBk would choose all three and the class assigned would be class A, but in IB1... is the choice of the first nearest neighbour a random choice? Is this the difference?
 
Thanks,
 
Santi.

_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist