I am doing semi-supervised classification with WEKA. The test set of my data (twitter) is unlabeled (no class assigned) but replaced with (?). I used WEKA to convert the test data from csv to arff file, which sets automatic datatype of 'string' to the class attribute. When I try to run it after creating the model with my train data it seems to give errors regarding the string datatype given to the unlabelled class attribute. My question is what datatype is suitable for the class attribute in '?' to avoid glitches?
Thank you, Eduardo. I think the third alternative is ideal for the problem. I will attend to it and get back.
As per my suggestion, instead of using '?' for unlabelled class use an assumption class. When u run any classification algorithm, the confusion matrix displays whether the assumption is true or false.
If assumption is true, then the predicted class of the test instance is the class mentioned, else the remaining class can be the apt prediction.
Hi ,
Any data type can have a label with "?". Attribute type with "?" value is most likely to be from a nominal type, hence, if you have such attribute but with string type (which is weird), simply apply "StringToNominal" filter. This will convert the string attribute into a nominal one after specifying the index of this particular attribute.
HTH.
Samer
- Badges
- Science topic