Skip to content

Ensemble round 2: AdaBoost, ExtraTrees, random split support #11

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
kundanjan merged 1 commit into from
Jun 30, 2026
Merged

Ensemble round 2: AdaBoost, ExtraTrees, random split support #11

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
370 changes: 370 additions & 0 deletions ensemble/src/main/java/org/sklearn/ensemble/AdaBoostClassifier.java
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,370 @@
package org.sklearn.ensemble;

import org.sklearn.core.Predictor;
import org.sklearn.math.Matrix;
import org.sklearn.math.RandomGenerator;
import org.sklearn.math.Vector;
import org.sklearn.tree.DecisionTreeClassifier;
import org.sklearn.utils.Validation;

import java.util.*;

/**
* AdaBoost classifier.
*
* <p>Fits a sequence of weak learners (default: decision stumps with
* {@code maxDepth=1}) on weighted versions of the training data. At each
* iteration sample weights are updated to give more weight to misclassified
* samples. Prediction uses weighted majority voting (SAMME algorithm).
*
* <p>Mirrors {@code sklearn.ensemble.AdaBoostClassifier}.
*
* <p>Usage:
* <pre>{@code
* AdaBoostClassifier ada = new AdaBoostClassifier(50, 1, 42);
* ada.fit(X, y);
* Vector preds = ada.predict(X_test);
* }</pre>
*/
public class AdaBoostClassifier implements Predictor<Matrix, Vector, Vector> {

private int nEstimators;
private int maxDepth;
private double learningRate;
private long seed;
private boolean fitted;
private List<DecisionTreeClassifier> estimators;
private double[] estimatorWeights;
private int[] classes;
private int nFeatures;

/**
* Create an AdaBoost classifier.
*
* @param nEstimators number of weak learners
* @param maxDepth max depth of each weak learner (default 1 = stump)
* @param seed random seed
*/
public AdaBoostClassifier(int nEstimators, int maxDepth, long seed) {
this(nEstimators, maxDepth, 1.0, seed);
}

/**
* Create an AdaBoost classifier with full control.
*
* @param nEstimators number of weak learners
* @param maxDepth max depth of each weak learner
* @param learningRate learning rate (shrinks contribution of each learner)
* @param seed random seed
*/
public AdaBoostClassifier(int nEstimators, int maxDepth,
double learningRate, long seed) {
if (nEstimators < 1) {
throw new IllegalArgumentException("nEstimators must be >= 1");
}
if (maxDepth < 1) {
throw new IllegalArgumentException("maxDepth must be >= 1");
}
if (learningRate <= 0) {
throw new IllegalArgumentException("learningRate must be > 0");
}
this.nEstimators = nEstimators;
this.maxDepth = maxDepth;
this.learningRate = learningRate;
this.seed = seed;
}

@Override
public AdaBoostClassifier fit(Matrix X, Vector y) {
Validation.checkMatrix(X, -1);
Validation.checkTarget(y, X.rows());

int n = X.rows();
int m = X.cols();
this.nFeatures = m;

Set<Integer> uniqueLabels = new LinkedHashSet<>();
for (int i = 0; i < n; i++) {
uniqueLabels.add((int) y.get(i));
}
this.classes = uniqueLabels.stream().mapToInt(Integer::intValue).toArray();
Arrays.sort(this.classes);
int nClasses = classes.length;

double[] sampleWeights = new double[n];
Arrays.fill(sampleWeights, 1.0 / n);

estimators = new ArrayList<>();
estimatorWeights = new double[nEstimators];
double epsilon = 1e-10;

for (int t = 0; t < nEstimators; t++) {
for (int i = 0; i < n; i++) {
if (sampleWeights[i] == 0) {
sampleWeights[i] = epsilon;
}
}

Object[] bootData = bootstrapSample(X, y, sampleWeights, t);
DecisionTreeClassifier stump = new DecisionTreeClassifier(
maxDepth, 2, 1, "gini");
stump.fit((Matrix) bootData[0], (Vector) bootData[1]);

Vector preds = stump.predict(X);
double error = 0.0;
for (int i = 0; i < n; i++) {
if (Math.abs(preds.get(i) - y.get(i)) >= 0.5) {
error += sampleWeights[i];
}
}
error /= sum(sampleWeights);

if (error > 1.0 - 1.0 / nClasses) {
estimatorWeights[t] = 0.0;
estimators.add(stump);
continue;
}

double alpha;
if (nClasses == 2) {
alpha = learningRate * 0.5 * Math.log((1 - error) / Math.max(error, epsilon));
} else {
alpha = learningRate * (Math.log((1 - error) / Math.max(error, epsilon))
+ Math.log(nClasses - 1));
}

estimatorWeights[t] = alpha;
estimators.add(stump);

double sumW = 0.0;
for (int i = 0; i < n; i++) {
boolean misclassified = Math.abs(preds.get(i) - y.get(i)) >= 0.5;
sampleWeights[i] *= Math.exp(alpha * (misclassified ? 1 : -1));
sampleWeights[i] = Math.max(sampleWeights[i], epsilon);
sumW += sampleWeights[i];
}
for (int i = 0; i < n; i++) {
sampleWeights[i] /= sumW;
}

if (error == 0) {
break;
}
}

fitted = true;
return this;
}

private Object[] bootstrapSample(Matrix X, Vector y, double[] weights, long iterSeed) {
int n = X.rows();
int m = X.cols();
RandomGenerator rng = new RandomGenerator(seed + iterSeed * 1000);

double[] cumSum = new double[n];
cumSum[0] = weights[0];
for (int i = 1; i < n; i++) {
cumSum[i] = cumSum[i - 1] + weights[i];
}
double totalW = cumSum[n - 1];

Matrix bootX = new Matrix(n, m);
Vector bootY = new Vector(n);
for (int i = 0; i < n; i++) {
double r = rng.nextDouble() * totalW;
int idx = lowerBound(cumSum, r);
idx = Math.min(idx, n - 1);
for (int j = 0; j < m; j++) {
bootX.set(i, j, X.get(idx, j));
}
bootY.set(i, y.get(idx));
}
return new Object[]{bootX, bootY};
}

private int lowerBound(double[] arr, double val) {
int lo = 0, hi = arr.length;
while (lo < hi) {
int mid = (lo + hi) / 2;
if (arr[mid] < val) {
lo = mid + 1;
} else {
hi = mid;
}
}
return lo;
}

private double sum(double[] arr) {
double s = 0;
for (double v : arr) {
s += v;
}
return s;
}

@Override
public Vector predict(Matrix X) {
Validation.checkFitted(fitted, "AdaBoostClassifier");
Validation.checkMatrix(X, -1);
if (X.cols() != nFeatures) {
throw new IllegalArgumentException(
"Feature dimension mismatch: expected " + nFeatures
+ " features, got " + X.cols());
}

int n = X.rows();
int nClasses = classes.length;

if (nClasses == 2) {
double[] weightedScores = new double[n];
for (int t = 0; t < estimators.size(); t++) {
if (estimatorWeights[t] == 0) {
continue;
}
Vector preds = estimators.get(t).predict(X);
for (int i = 0; i < n; i++) {
weightedScores[i] += estimatorWeights[t]
* (preds.get(i) == classes[1] ? 1 : -1);
}
}
double[] out = new double[n];
for (int i = 0; i < n; i++) {
out[i] = weightedScores[i] >= 0 ? classes[1] : classes[0];
}
return new Vector(out);
} else {
double[][] votes = new double[n][nClasses];
for (int t = 0; t < estimators.size(); t++) {
if (estimatorWeights[t] == 0) {
continue;
}
Vector preds = estimators.get(t).predict(X);
for (int i = 0; i < n; i++) {
int predClass = (int) preds.get(i);
int idx = classIndex(predClass);
votes[i][idx] += estimatorWeights[t];
}
}
double[] out = new double[n];
for (int i = 0; i < n; i++) {
out[i] = classes[argmax(votes[i])];
}
return new Vector(out);
}
}

/**
* Predict class probabilities based on normalized weighted votes.
*/
public Vector predictProba(Matrix X) {
Validation.checkFitted(fitted, "AdaBoostClassifier");
Validation.checkMatrix(X, -1);

int n = X.rows();
int nClasses = classes.length;

double[] probs = new double[n];

if (nClasses == 2) {
double maxAbsWeight = 0;
double[] scores = new double[n];
for (int t = 0; t < estimators.size(); t++) {
if (estimatorWeights[t] == 0) {
continue;
}
Vector preds = estimators.get(t).predict(X);
double absW = Math.abs(estimatorWeights[t]);
maxAbsWeight += absW;
for (int i = 0; i < n; i++) {
scores[i] += estimatorWeights[t]
* (preds.get(i) == classes[1] ? 1 : -1);
}
}
for (int i = 0; i < n; i++) {
probs[i] = maxAbsWeight > 0
? (scores[i] / maxAbsWeight + 1.0) / 2.0
: 0.5;
probs[i] = Math.max(0.0, Math.min(1.0, probs[i]));
}
} else {
double[][] votes = new double[n][nClasses];
for (int t = 0; t < estimators.size(); t++) {
if (estimatorWeights[t] == 0) {
continue;
}
Vector preds = estimators.get(t).predict(X);
for (int i = 0; i < n; i++) {
int idx = classIndex((int) preds.get(i));
votes[i][idx] += estimatorWeights[t];
}
}
for (int i = 0; i < n; i++) {
double minVote = Double.MAX_VALUE, maxVote = -Double.MAX_VALUE;
for (int k = 0; k < nClasses; k++) {
if (votes[i][k] < minVote) {
minVote = votes[i][k];
}
if (votes[i][k] > maxVote) {
maxVote = votes[i][k];
}
}
double range = maxVote - minVote;
if (range > 0) {
probs[i] = (votes[i][classIndex(classes[1])] - minVote) / range;
} else {
probs[i] = 0.5;
}
}
}
return new Vector(probs);
}

@Override
public double score(Matrix X, Vector y) {
Validation.checkFitted(fitted, "AdaBoostClassifier");
Validation.checkMatrix(X, -1);
Validation.checkTarget(y, X.rows());

Vector pred = predict(X);
int correct = 0;
for (int i = 0; i < y.size(); i++) {
if (Math.abs(pred.get(i) - y.get(i)) < 0.5) {
correct++;
}
}
return (double) correct / y.size();
}

private int classIndex(int cls) {
for (int i = 0; i < classes.length; i++) {
if (classes[i] == cls) {
return i;
}
}
return 0;
}

private int argmax(double[] arr) {
int best = 0;
for (int i = 1; i < arr.length; i++) {
if (arr[i] > arr[best]) {
best = i;
}
}
return best;
}

public boolean isFitted() {
return fitted;
}

@Override
public Map<String, Object> getParameters() {
Map<String, Object> params = new LinkedHashMap<>();
params.put("n_estimators", nEstimators);
params.put("max_depth", maxDepth);
params.put("learning_rate", learningRate);
return Collections.unmodifiableMap(params);
}
}
Loading
Loading