Transform method

The transform method alows user to convert original dataframe into it’s numerical representation. Thanks to that, we can use our metric with external libraries such as scipy or scikit-learn. More on that in External Api Compatibility section.

Transform

It is done by calling the transform method on a fitted Gower instance.

import numpy as np

from gower_metric import Gower

data = np.array([[1, 'a'], [2, 'b'], [3, 'a'], [4, 'c']], dtype=object)

feature_types = {
   0: "ratio_scale_interval",
   1: "categorical_nominal",
}

gower = Gower(feature_types=feature_types).fit(data)

transformed_data = gower.transform(data)

All numerical, ratio scale interval features are intact. Binary features are represented as 0 and 1, depending on the value. Categorical nominal and ordinal features are encoded using ordinal encoding from scikit-learn.

Fit transform

For convenience, Gower also implements fit_transform method, which combines fit and transform in one call.

import numpy as np

from gower_metric import Gower

data = np.array([[1, 'a'], [2, 'b'], [3, 'a'], [4, 'c']], dtype=object)

feature_types = {
   0: "ratio_scale_interval",
   1: "categorical_nominal",
}

gower = Gower(feature_types=feature_types)

transformed_data = gower.fit_transform(data)

Warning

Under the hood, the fit method learns the mapping of categorical ordinal values to their numerical representation. Therefore, calling transform on the same data before and after re-fitting the instance may result in different numerical representations or even NaN values. The same applies to the fit_transform method. If your data does not contain any categorical ordinal features, this warning may not apply (we have not implemented tests for this scenario).