[SOLVED] Using sklearn for hierarchical classification

Using sklearn for hierarchical classification

I was wondering if hierarchical classifications are supported by the sciki-learn library. I am dealing with the 3 classes divided by 6 subclasses each, such as:

import numpy as np
from sklearn.tree import DecisionTreeClassifier

X = np.random.randn(5, 1)

number, rows, cols = 5, 3, 6
y = np.zeros((number, rows, cols), dtype=int)
for n in range(number):
    for row in range(rows):
        col = np.random.randint(cols)
        y[n, row, col] = 1

tree = DecisionTreeClassifier()
tree.fit(X, y)

but find error:

ValueError: Found array with dim 3. DecisionTreeClassifier expected <= 2.

Solution

You can use hiclass.

pip install hiclass

Train the model:

from sklearn.tree import DecisionTreeClassifier
from hiclass.MultiLabelLocalClassifierPerNode import MultiLabelLocalClassifierPerNode

tree = DecisionTreeClassifier()
classifier = MultiLabelLocalClassifierPerNode(local_classifier=tree)
classifier.fit(X, y)

Test and measure precision:

hiclass.metrics import precision

predictions = classifier.predict(y_test)
p = precision(y_test, predictions)
print(p)

Refer to the hiclass paper for complementary information.