Hello all - in particular to anyone who has experience with
Classification Trees.
I have found an odd discrepancy in Classification Trees that is giving
me a headache.
1. I'm using the TREE command, with /CHAID INTERVALS=10. This
makes scale variables to be subdivided into 10 approximately equally
sized groups before analysis for significant splits. Looking at the
resultant tree diagram it is applying rules for scale variables with 3
significant digits e.g. 'V322<=0.087'
2. I'm then trying to replicate the result using the rule set
generated by the /RULES subcommand. These rules deal with the scale
variables with many significant digits e.g. 'V322<=0.0869999999999999'
Because some of my cases have a value for V322=0.087 exactly, they are
being treated differently according to whether I run the TREE command
(1) or the syntax rules generated at the end of it (2). I'm therefore
getting a different final result at the end.
Has anyone seen this before? Is it a known issue?
Are there any workarounds? (I'm currently looking at processing the raw
text of the syntax rules and reducing the number of significant digits.)
Thanks,
Mark
====================To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
