Closed danspeck closed 1 week ago
I'm not familiar with MULAN, so not sure which characters it doesn't like. However, I've added a -no-rename
command-line option (in the GUI it is the inverse: renameAttributes
), which allows you to turn off the attribute renaming. It may or may not work with your dataset. ;-)
Have you tried a new snapshot with the new functionality?
First, thank you for your speedy response! I really didn't expect to get such a quick response to my issue. I have not yet built the latest branch but will try to do so soon. I initially had trouble building master because of some unresolved dependencies so I worked around the issue by building a map from the original attribute offsets to the original attribute names and then "fixed" the data after loading it.
I appreciate your work on this project and will let you know how the new option works.
Also, I'm not using the Meka program proper but am attempting to integrate its libraries into a separate program that does multilabel classification.
What unresolved dependencies did you encounter? The master branch gets built every night and a snapshot made available.
I am no longer getting the same messages I got when I originally tried to compile meka from the master branch. I noticed that there were several changes made in July and I may have grabbed a version that had an issue that no longer exists. I believe the issue was with a dependency coming from the pentaho-public repo but I can't recall exactly which dependency.
Great to hear that the build problem disappeared!
I've recently started working with Meka and Weka and ran into an issue with the renaming of input attributes in MULAN.java, specifically: https://github.com/Waikato/meka/blame/master/src/main/java/meka/classifiers/multilabel/MULAN.java#L139
In order to use a classifier that has been built using MULAN, I have to map my original input attributes to the renamed (a_N) and resequenced ones. I'm wondering if it wouldn't be better to just make the attributes conform to MULAN's requirements, e.g., by replacing any offensive characters (e.g., '-') by inoffensive ones. That would make applying the generated classifiers easier to use.