"Weka J48 and Rapidminer W-J48 Different"

martyns

Using the same dataset I ran WEKAs implementation of J-48 and then Rapidminers implementation of WEKAs J-48. WJ-48.

It could be expected that using the same dataset the same tree would be constructed but it is not.

Some parts are very nearly the same but differ but a small amount on a split or percent but other parts are structurally different.

I checked that the same default internal parameters were being used.

I have included two attachments visualising the same section of both trees.
Thanks!



[attachment deleted by admin]

haddock

I'm not entirely sure that RapidMiner actually implement the WEKA operators, perhaps they just use the library, in which case you need to check that the RM version is exactly the same version as the WEKA version.