05-25-2015 10:44 AM
After run 'rule induction' in EM12.1, from the results I got the first rip with a 100 pure node(say node 7) , but the next rips didn't take the 7 node rows away. Anyone knows why? Thanks.
05-29-2015 01:37 PM
Actual situation was that node 4, in rip4, had 100% pure (see below a tested sample), but was not removed (it said: no leaf was rippd from the model), I expected node 4's rows/observations should be removed for next rips, or further modeling process.
RIP4 Leaf Table: Threshold= 100
No leaf was ripped from the model.
Node N target=0 target=1
4 484942 1.0000 0.0000
11 50101 0.9993 0.0007
05-29-2015 02:00 PM
I see, thank you for clarifying.
My only thought is that with the large sample size for node 4, maybe the purity isn't exactly 100% as shown, but actually 99.999 or something and that precision is just not displayed here. You could try changing the Purity Threshold property to 99 and see if it rips node 4 as expected in that case.