AdaBoost - アンサンブル手法、分類: 教師あり機械学習

Boosting

Definition and Purpose

Boosting is an ensemble learning technique used in machine learning to improve the accuracy of models. It combines multiple weak classifiers (models that perform slightly better than random guessing) to create a strong classifier. The main purpose of boosting is to sequentially apply the weak classifiers to the data, correcting the errors made by the previous classifiers, and thus improve overall performance.

Key Objectives:

• Improve Accuracy: Enhance the prediction accuracy by combining the outputs of several weak classifiers.
• Reduce Bias and Variance: Address issues of bias and variance to achieve a better generalization of the model.
• Handle Complex Data: Effectively model complex relationships in the data.

AdaBoost (Adaptive Boosting)

Definition and Purpose

AdaBoost, short for Adaptive Boosting, is a popular boosting algorithm. It adjusts the weights of incorrectly classified instances so that subsequent classifiers focus more on difficult cases. The main purpose of AdaBoost is to improve the performance of weak classifiers by emphasizing the hard-to-classify examples in each iteration.

Key Objectives:

• Weight Adjustment: Increase the weight of misclassified instances to ensure the next classifier focuses on them.
• Sequential Learning: Build classifiers sequentially, where each new classifier corrects the errors of its predecessor.
• Improved Performance: Combine weak classifiers to form a strong classifier with better predictive power.

How AdaBoost Works

1. Initialize Weights:

   • Assign equal weights to all training instances. For a dataset with n instances, each instance has a weight of 1/n.

2. Train Weak Classifier:

   • Train a weak classifier using the weighted dataset.

3. Calculate Classifier Error:

   • Compute the error of the weak classifier, which is the sum of the weights of misclassified instances.

4. Compute Classifier Weight:

   • Calculate the weight of the classifier based on its error. The weight is given by: alpha = 0.5 * log((1 - error) / error)
   • A lower error results in a higher classifier weight.

5. Update Weights of Instances:

   • Adjust the weights of the instances. Increase the weights of misclassified instances and decrease the weights of correctly classified instances.
   • The updated weight for instance i is: weight[i] = weight[i] * exp(alpha * (misclassified ? 1 : -1))
   • Normalize the weights to ensure they sum to 1.

6. Combine Weak Classifiers:

   • The final strong classifier is a weighted sum of the weak classifiers: Final classifier = sign(sum(alpha * weak_classifier))
   • The sign function determines the class label based on the sum.

AdaBoost (Binary Classification) Example

AdaBoost, short for Adaptive Boosting, is an ensemble technique that combines multiple weak classifiers to create a strong classifier. This example demonstrates how to implement AdaBoost for binary classification using synthetic data, evaluate the model's performance, and visualize the decision boundary.

Python Code Example

1. Import Libraries

import numpy as np
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from sklearn.ensemble import AdaBoostClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.metrics import accuracy_score, confusion_matrix, classification_report

This block imports the necessary libraries for data manipulation, plotting, and machine learning.

2. Generate Sample Data

np.random.seed(42)  # For reproducibility

# Generate synthetic data for 2 classes
n_samples = 1000
n_samples_per_class = n_samples // 2

# Class 0: Centered around (-1, -1)
X0 = np.random.randn(n_samples_per_class, 2) * 0.7 + [-1, -1]

# Class 1: Centered around (1, 1)
X1 = np.random.randn(n_samples_per_class, 2) * 0.7 + [1, 1]

# Combine the data
X = np.vstack([X0, X1])
y = np.hstack([np.zeros(n_samples_per_class), np.ones(n_samples_per_class)])

# Shuffle the dataset
shuffle_idx = np.random.permutation(n_samples)
X, y = X[shuffle_idx], y[shuffle_idx]

This block generates synthetic data with two features, where the target variable y is defined based on the class center, simulating a binary classification scenario.

3. Split the Dataset

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

This block splits the dataset into training and testing sets for model evaluation.

4. Create and Train the AdaBoost Classifier

base_estimator = DecisionTreeClassifier(max_depth=1)  # Decision stump
model = AdaBoostClassifier(estimator=base_estimator, n_estimators=3, random_state=42)
model.fit(X_train, y_train)

This block initializes the AdaBoost model with a decision stump as the base estimator and trains it using the training dataset.

5. Make Predictions

y_pred = model.predict(X_test)

This block uses the trained model to make predictions on the test set.

6. Evaluate the Model

accuracy = accuracy_score(y_test, y_pred)
conf_matrix = confusion_matrix(y_test, y_pred)
class_report = classification_report(y_test, y_pred)

print(f"Accuracy: {accuracy:.4f}")
print("\nConfusion Matrix:")
print(conf_matrix)
print("\nClassification Report:")
print(class_report)

Output:

Accuracy: 0.9400

Confusion Matrix:
[[96  8]
 [ 4 92]]

Classification Report:
              precision    recall  f1-score   support

         0.0       0.96      0.92      0.94       104
         1.0       0.92      0.96      0.94        96

    accuracy                           0.94       200
   macro avg       0.94      0.94      0.94       200
weighted avg       0.94      0.94      0.94       200

This block calculates and prints the accuracy, confusion matrix, and classification report, providing insights into the model's performance.

7. Visualize the Decision Boundary

x_min, x_max = X[:, 0].min() - 1, X[:, 0].max() + 1
y_min, y_max = X[:, 1].min() - 1, X[:, 1].max() + 1
xx, yy = np.meshgrid(np.arange(x_min, x_max, 0.1),
                     np.arange(y_min, y_max, 0.1))
Z = model.predict(np.c_[xx.ravel(), yy.ravel()])
Z = Z.reshape(xx.shape)

plt.figure(figsize=(10, 8))
plt.contourf(xx, yy, Z, alpha=0.4, cmap='RdYlBu')
scatter = plt.scatter(X[:, 0], X[:, 1], c=y, cmap='RdYlBu', edgecolor='black')
plt.xlabel("Feature 1")
plt.ylabel("Feature 2")
plt.title("AdaBoost Binary Classification")
plt.colorbar(scatter)
plt.show()

This block visualizes the decision boundary created by the AdaBoost model, illustrating how the model separates the two classes in the feature space.

Output:

This structured approach demonstrates how to implement and evaluate AdaBoost for binary classification tasks, providing a clear understanding of its capabilities. The visualization of the decision boundary aids in interpreting the model's predictions.

AdaBoost (Multiclass Classification) Example

AdaBoost is an ensemble learning technique that combines multiple weak classifiers to create a strong classifier. This example demonstrates how to implement AdaBoost for multiclass classification using synthetic data, evaluate the model's performance, and visualize the decision boundary for five classes.

Python Code Example

1. Import Libraries

import numpy as np
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from sklearn.ensemble import AdaBoostClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.metrics import accuracy_score, confusion_matrix, classification_report

This block imports the necessary libraries for data manipulation, plotting, and machine learning.

2. Generate Sample Data with 5 Classes

np.random.seed(42)  # For reproducibility
n_samples = 2500  # Total number of samples
n_samples_per_class = n_samples // 5  # Ensure this is exactly n_samples // 5

# Class 0: Centered around (-2, -2)
X0 = np.random.randn(n_samples_per_class, 2) * 0.5 + [-2, -2]

# Class 1: Centered around (0, -2)
X1 = np.random.randn(n_samples_per_class, 2) * 0.5 + [0, -2]

# Class 2: Centered around (2, -2)
X2 = np.random.randn(n_samples_per_class, 2) * 0.5 + [2, -2]

# Class 3: Centered around (-1, 2)
X3 = np.random.randn(n_samples_per_class, 2) * 0.5 + [-1, 2]

# Class 4: Centered around (1, 2)
X4 = np.random.randn(n_samples_per_class, 2) * 0.5 + [1, 2]

# Combine the data
X = np.vstack([X0, X1, X2, X3, X4])
y = np.hstack([np.zeros(n_samples_per_class), 
               np.ones(n_samples_per_class),
               np.full(n_samples_per_class, 2),
               np.full(n_samples_per_class, 3),
               np.full(n_samples_per_class, 4)])

# Shuffle the dataset
shuffle_idx = np.random.permutation(n_samples)
X, y = X[shuffle_idx], y[shuffle_idx]

This block generates synthetic data for five classes located in different regions of the feature space.

3. Split the Dataset

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

This block splits the dataset into training and testing sets for model evaluation.

4. Create and Train the AdaBoost Classifier

base_estimator = DecisionTreeClassifier(max_depth=1)  # Decision stump
model = AdaBoostClassifier(estimator=base_estimator, n_estimators=10, random_state=42)
model.fit(X_train, y_train)

This block initializes the AdaBoost classifier with a weak learner (decision stump) and trains it using the training dataset.

5. Make Predictions

y_pred = model.predict(X_test)

This block uses the trained model to make predictions on the test set.

6. Evaluate the Model

accuracy = accuracy_score(y_test, y_pred)
conf_matrix = confusion_matrix(y_test, y_pred)
class_report = classification_report(y_test, y_pred)

print(f"Accuracy: {accuracy:.4f}")
print("\nConfusion Matrix:")
print(conf_matrix)
print("\nClassification Report:")
print(class_report)

Output:

Accuracy: 0.9540

Confusion Matrix:
[[ 97   2   0   0   0]
 [  0  92   3   0   0]
 [  0   4  92   0   0]
 [  0   0   0  86  14]
 [  0   0   0   0 110]]

Classification Report:
              precision    recall  f1-score   support

         0.0       1.00      0.98      0.99        99
         1.0       0.94      0.97      0.95        95
         2.0       0.97      0.96      0.96        96
         3.0       1.00      0.86      0.92       100
         4.0       0.89      1.00      0.94       110

    accuracy                           0.95       500
   macro avg       0.96      0.95      0.95       500
weighted avg       0.96      0.95      0.95       500

This block calculates and prints the accuracy, confusion matrix, and classification report, providing insights into the model's performance.

7. Visualize the Decision Boundary

x_min, x_max = X[:, 0].min() - 1, X[:, 0].max() + 1
y_min, y_max = X[:, 1].min() - 1, X[:, 1].max() + 1
xx, yy = np.meshgrid(np.arange(x_min, x_max, 0.1),
                     np.arange(y_min, y_max, 0.1))
Z = model.predict(np.c_[xx.ravel(), yy.ravel()])
Z = Z.reshape(xx.shape)

plt.figure(figsize=(12, 10))
plt.contourf(xx, yy, Z, alpha=0.4, cmap='viridis')
scatter = plt.scatter(X[:, 0], X[:, 1], c=y, cmap='viridis', edgecolor='black')
plt.xlabel("Feature 1")
plt.ylabel("Feature 2")
plt.title("AdaBoost Multiclass Classification (5 Classes)")
plt.colorbar(scatter)
plt.show()

This block visualizes the decision boundaries created by the AdaBoost classifier, illustrating how the model separates the five classes in the feature space.

Output:

This structured approach demonstrates how to implement and evaluate AdaBoost for multiclass classification tasks, providing a clear understanding of its capabilities and the effectiveness of visualizing decision boundaries.

以上がAdaBoost - アンサンブル手法、分類: 教師あり機械学習の詳細内容です。詳細については、PHP 中国語 Web サイトの他の関連記事を参照してください。