Thank you dear Martin for this explanation.
I want to know that whether or not the unlabeled data improves prediction
Unlabeled data with label data affect on creating prediction model. In
addition, the way you prepare your data, helps to provide reliable
Data set has 206 instances. It consists of the labeled data (54
unlabeled data (206 instances). I split the labeled data
into training/test set according to percentage. 67% is training set (36
instances) and 33% is test set (18 instances). In the experiment, i used
training and test set to classify with J48. Then, i used the unlabeled data
to classify with YATSI.
Are the files and process correct?
Not completely correct.
Your files can be accepted. The aim of semi-supervised learning is that:
labels are expensive and we usually have only small amount of it. However,
it's easy to get unlabeled data. By using label and unlabeled data we can
build the ultimate model.
Perform semi-supervised learning using YATSI from the "Collective" panel.
Create the model there, save it and load it in the "Classify" panel. After
that you can use either J48 or any other scheme to perform classification
from Classify panel.
I'm wondering about design structure of my test file. The cause of
that the instances in test file must come from a training set
only or not? (your attached new test file come from a training set)
Test set is something that you wouldn't have in the real-world conditions.
We just use this set to get the statistics in order to assess the
performance of our model. Obtaining this set from the training file helps
in the evaluation stage. You had an evaluation problem because instances of
the test set could not be found.
On 22 Jul 2015 03:27, "Beam [via WEKA]" <
To do the classification with 'YATSI
algorithm', I loaded the training set
to the Preprocess tab. Next, from the Collective tab, I chose the
'unlabeled/Test set' option and loaded each file into Unlabeled set and
Test set respectively. -> RUN
after that I am getting this problem.
How should i do? Can anybody help me?
Thanks in advance
If you reply to this email, your message will be added to the discussion
To unsubscribe from WEKA, click here
View this message in context:
Sent from the WEKA mailing list archive at Nabble.com