Open bubbazz opened 2 years ago
The unlabeled dataset requires the exact same structure as the training set (ie same attribute and nominal label order) and the class attribute columns to contain only missing values (ie ?
).
If you need to introduce missing values, have a look at the missing-values-imputation Weka package.
I've added a note to Tutorial.tex to make it clearer. Thanks for pointing it out!
Thanks for clearing it up. it helped me a lot.
Dear Meka-Team,
Is it possible to combine semi-supervised learning with hyperparameter tuning?
after training and testing (the two seperate commands), how do you predict unseen data.
Thank you very much indeed.
With kind regards
From a quick look at the code:
distributionForInstance
method for each row in the weka.core.Instance
object) using the meka.core.ThresholdUtils class.Please note, I don't use Meka, so only some vague pointers.
Hi,
in the tutorial under the paragraph semi-supervised learning. is in the command an unlabed.arff.
I wonder how in the arff such a line looks. The only thing I found are "?" as attribute values.
For example: @RELATION unlabeded @ATTRIBUTE X1 NUMERIC @ATTRIBUTE X2 NUMERIC @ATTRIBUTE y0 {0, 1} @ATTRIBUTE y1 {0, 1}
{ 0 42.42, 1 42.42, 2 ?, 3 ? }
Is the above described unlabed? or what does such a dataset look like?