Preserving feature names about sklearn2pmml

[vruusmann]

The RandomForestClassifier object does not contain feature names. If you want to override default names (eg. "x1", "x2", "x3"), then you have to construct and pass a DataFrameMapper object to the sklearn2pmml method.

You can create a dummy DataFrameMapper object (that specifies the names of thousand columns) using a single line of Python application code. That's not overkill.

Or are you suggesting that "x1", "x2", etc. should be renamed to "feature1", "feature2", etc?

[vruusmann]

@KidCrippler I'm not a Python expert either, so I'm going to construct the DataFrameMapper object using four lines of Python application code (instead of one as advertised above). The idea is to map a list of strings to a None transformation:

iris_active_fields = ["Sepal.Length", "Sepal.Width", "Petal.Length", "Petal.Width"]
iris_target_field = "Species"

iris_mapper = sklearn_pandas.DataFrameMapper([
    (iris_active_fields, None),
    (iris_target_field, None)
])

sklearn2pmml(iris_classifier, iris_mapper, "Iris.pmml")

[KidCrippler]

@vruusmann I had the exact same problem.
I'm not a very proficient python programmer and couldn't get around wrapping my model with an "identity" mapper that only modifies the predictor names from x1, x2, etc... to whatever I want them to be.
Can you please provide a more detailed explanation on how to do that?

10x