How to set up and use sample weight in the Orange python package?

I am new to the Orange python package for data mining. I am using Orange 2.7.

My dataset has a binary target (Good and Bad). The Good instances are down sampled with a sampling weight of 10. How can I set up and use the weight for classfication analysis in both Windows and Linux versions of Orange? Thank you for your help!

Solution

You have to add a new meta column to your data, containing the instance weights (see Meta attributes and Table.add_meta_attribute. Store the meta column's id and call the learner with that meta id.

import Orange
iris = Orange.data.Table("iris")
# Add some weights to the iris dataset
weight = Orange.feature.Continuous("weight")
weight_id = -10
iris.domain.add_meta(weight_id, weight)
iris.add_meta_attribute(weight, 1.0)
for i in range(50, 150):
     iris[i][weight] = 10

# Train a tree classifier on weighted data.
clsf = Orange.classification.tree.TreeLearner(iris, weight_id)

# Evaluate learner performance on weighted data
results = Orange.evaluation.testing.cross_validation(
    [Orange.classification.tree.TreeLearner,
     Orange.classification.bayes.NaiveLearner],
    (iris, weight_id)  # Note how you pass the weight id to testing functions
)
auc = Orange.evaluation.scoring.AUC(results)
ca = Orange.evaluation.scoring.CA(results)