Multiclass and Multioutput Algorithms

Introduction

In this lab, we will explore the functionality and usage of multiclass and multioutput algorithms in scikit-learn. Multiclass classification is a classification task where samples are assigned to more than two classes. Multioutput classification, on the other hand, predicts multiple properties for each sample. We will cover the following topics:

1. Multiclass Classification
2. Multilabel Classification
3. Multiclass-Multioutput Classification
4. Multioutput Regression

VM Tips

After the VM startup is done, click the top left corner to switch to the Notebook tab to access Jupyter Notebook for practice.

Sometimes, you may need to wait a few seconds for Jupyter Notebook to finish loading. The validation of operations cannot be automated because of limitations in Jupyter Notebook.

If you face issues during learning, feel free to ask Labby. Provide feedback after the session, and we will promptly resolve the problem for you.

Skills Graph

Multiclass Classification

Problem Description

Multiclass classification is a classification task with more than two classes. Each sample is assigned to only one class.

Target Format

A valid representation of multiclass targets is a 1D or column vector containing more than two discrete values.

Example

Let's use the Iris dataset to demonstrate multiclass classification:

from sklearn import datasets
from sklearn.linear_model import LogisticRegression
from sklearn.multiclass import OneVsRestClassifier

## Load the Iris dataset
X, y = datasets.load_iris(return_X_y=True)

## Fit a logistic regression model using OneVsRestClassifier
model = OneVsRestClassifier(LogisticRegression())
model.fit(X, y)

## Make predictions
predictions = model.predict(X)
print(predictions)

Multilabel Classification

Problem Description

Multilabel classification is a classification task where each sample can be assigned multiple labels. The number of labels each sample can have is greater than two.

Target Format

A valid representation of multilabel targets is a binary matrix, where each row represents a sample and each column represents a class. A value of 1 indicates the presence of the label in the sample, while 0 or -1 indicates the absence.

Example

Let's create a multilabel classification problem using the make_classification function:

from sklearn.datasets import make_classification
from sklearn.multioutput import MultiOutputClassifier
from sklearn.ensemble import RandomForestClassifier

## Generate a multilabel classification problem
X, y = make_classification(n_samples=100, n_features=10, n_informative=5, random_state=0)
y = y.reshape(-1, 1)

## Fit a multioutput random forest classifier
model = MultiOutputClassifier(RandomForestClassifier())
model.fit(X, y)

## Make predictions
predictions = model.predict(X)
print(predictions)

Multiclass-Multioutput Classification

Problem Description

Multiclass-multioutput classification, also known as multitask classification, predicts multiple non-binary properties for each sample. Each property can have more than two classes.

Target Format

A valid representation of multiclass-multioutput targets is a dense matrix, where each row represents a sample and each column represents a different property or class.

Example

Let's create a multiclass-multioutput classification problem using the make_classification function:

from sklearn.datasets import make_classification
from sklearn.multioutput import MultiOutputClassifier
from sklearn.svm import SVC

## Generate a multiclass-multioutput classification problem
X, y = make_classification(n_samples=100, n_features=10, n_informative=5, n_classes=3, random_state=0)

## Fit a multioutput support vector classifier
model = MultiOutputClassifier(SVC())
model.fit(X, y)

## Make predictions
predictions = model.predict(X)
print(predictions)

Multioutput Regression

Problem Description

Multioutput regression predicts multiple numerical properties for each sample. Each property is a numerical variable, and the number of properties can be greater than or equal to two.

Target Format

A valid representation of multioutput regression targets is a dense matrix, where each row represents a sample and each column represents a different property.

Example

Let's create a multioutput regression problem using the make_regression function:

from sklearn.datasets import make_regression
from sklearn.multioutput import MultiOutputRegressor
from sklearn.linear_model import LinearRegression

## Generate a multioutput regression problem
X, y = make_regression(n_samples=100, n_features=10, n_targets=3, random_state=0)

## Fit a multioutput linear regression model
model = MultiOutputRegressor(LinearRegression())
model.fit(X, y)

## Make predictions
predictions = model.predict(X)
print(predictions)

Summary

In this lab, we explored multiclass and multioutput algorithms in scikit-learn. We covered multiclass classification, multilabel classification, multiclass-multioutput classification, and multioutput regression. These algorithms allow us to solve complex classification and regression tasks with multiple targets or classes.