How to best get class mapping for pickled StackedModel

[JoergPeisker]

Hi,

I am trying to deploy a stackedmodel classifier with a few dozen classes. The exporting and importing into a common python script works like a charm but how do I actually map the class-label back to the precited class-indices? Is there something like .classes_ or .class_indices?

[markotoplak]

To get predictions, the most Orange-like way is to prepare your data for prediction in an Orange Table and then just calling it with the data, for example as stackedmodel(data). This returns class label indices, which you can then transform to values, like this:

import Orange
import Orange.classification
import Orange.ensembles.stack
iris = Orange.data.Table("iris")
lr = Orange.classification.LogisticRegressionLearner()
stack = Orange.ensembles.stack.StackedLearner([lr], aggregate=lr)
stackedmodel = stack(iris[:140])

# use the model
predictions = stackedmodel(iris[140:])
print(predictions)

# to get classes from indices
values = stackedmodel.domain.class_var.values
print([values[int(v)] for v in predictions])

Which returns

[2. 2. 2. 2. 2. 2. 2. 2. 2. 2.]
['Iris-virginica', 'Iris-virginica', 'Iris-virginica', 'Iris-virginica', 'Iris-virginica', 'Iris-virginica', 'Iris-virginica', 'Iris-virginica', 'Iris-virginica', 'Iris-virginica']

I hope I understood your question correctly.

[JoergPeisker]

Thanks for the fast reply and my apologize for the short question. This pretty much solves my problem :-)