Advanced examples

In this section we will show how to use weighting of features, calculating similarities instead of distances, Podani’s optimization and more.

Calculating similarities

Package provides an option to calculate similarities instead of distances.

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)

similarity = gower.similarity(data[0], data[1])

Passing weights

You can pass weights to features when calculating distances or similarities. This allows you to give more importance to certain features over others.

import numpy as np

from gower_metric import Gower

data = np.array([[1, 'a', 3.5], [2, 'b', 4.0], [3, 'a', 2.5], [4, 'c', 5.0]], dtype=object)

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

weights = {
   0: 0.5,
   1: 2.0,
   2: 1.0
}

gower = Gower(feature_types=feature_types, feature_weights=weights).fit(data)

Categorical ordinal example

Here things get a bit more tricky. When dealing with categorical ordinal data, we need to provide an additional mapping that defines the order of the categories.

import numpy as np

from gower_metric import Gower

data = np.array([
    [1, 'low', 3.5],
    [2, 'medium', 4.0],
    [3, 'high', 2.5],
    [4, 'medium', 5.0]
], dtype=object)

feature_types = {
    0: "ratio_scale_interval",
    1: "categorical_ordinal",
    2: "ratio_scale_interval"
}

ordinal_mappings = {
    1: ['low', 'medium', 'high']
}

gower = Gower(feature_types=feature_types, categorical_ordinal_values_order=ordinal_mappings)
gower.fit(data)

More class functionality

On top of the examples before, we can also play with other class functionalities, mostly for numerical data but not only.

import numpy as np

from gower_metric import Gower

data = np.array([
    [1, 'low', 3.5],
    [2, 'medium', 4.0],
    [3, 'high', 2.5],
    [4, 'medium', 5.0]
], dtype=object)

feature_types = {
    0: "ratio_scale_interval",
    1: "categorical_ordinal",
    2: "ratio_scale_interval"
}

ordinal_mappings = {
    1: ['low', 'medium', 'high']
}

gower = Gower(
    feature_types=feature_types,
    categorical_ordinal_values_order=ordinal_mappings,
    categorical_ordinal_calculation_type="podani",
    scale="iqr",
    missing_strategy="max_dist",
    scale_window="kde",
    scale_window_type="silverman",
    conditional_distances=True
)
gower.fit(data)