Advanced ML Models

1/**2 * Example 10: Advanced ML Models3 *4 * Five models on tiny datasets, each shown with its own fit/predict/score calls:5 * - KMeans clustering6 * - K-Nearest Neighbors (classification and regression)7 * - PCA (dimensionality reduction)8 * - Gaussian Naive Bayes9 */10 11import { accuracy } from "deepbox/metrics";12import { GaussianNB, KMeans, KNeighborsClassifier, KNeighborsRegressor, PCA } from "deepbox/ml";13import { tensor } from "deepbox/ndarray";14import { trainTestSplit } from "deepbox/preprocess";15 16console.log("=".repeat(60));17console.log("Example 10: Advanced ML Models");18console.log("=".repeat(60));19 20// ============================================================================21// Part 1: KMeans Clustering22// ============================================================================23console.log("\nPart 1: KMeans Clustering");24console.log("-".repeat(60));25 26const clusterData = tensor([27  [1, 2],28  [1.5, 1.8],29  [5, 8],30  [8, 8],31  [1, 0.6],32  [9, 11],33  [8, 2],34  [10, 2],35  [9, 3],36]);37 38const kmeans = new KMeans({ nClusters: 3, randomState: 42 });39kmeans.fit(clusterData);40 41const clusterLabels = kmeans.predict(clusterData);42console.log("Cluster labels:", clusterLabels.toString());43console.log("Cluster centers shape:", kmeans.clusterCenters.shape);44console.log("Inertia:", kmeans.inertia.toFixed(4));45console.log("Number of iterations:", kmeans.nIter);46 47// ============================================================================48// Part 2: K-Nearest Neighbors Classification49// ============================================================================50console.log("\nPart 2: K-Nearest Neighbors Classification");51console.log("-".repeat(60));52 53const XClass = tensor([54  [0, 0],55  [1, 1],56  [2, 2],57  [3, 3],58  [4, 4],59  [5, 5],60  [6, 6],61  [7, 7],62]);63const yClass = tensor([0, 0, 0, 0, 1, 1, 1, 1]);64 65const [XTrainKNN, XTestKNN, yTrainKNN, yTestKNN] = trainTestSplit(XClass, yClass, {66  testSize: 0.25,67  randomState: 42,68});69 70const knnClassifier = new KNeighborsClassifier({ nNeighbors: 3 });71knnClassifier.fit(XTrainKNN, yTrainKNN);72 73const yPredKNN = knnClassifier.predict(XTestKNN);74const knnAccuracy = accuracy(yTestKNN, yPredKNN);75 76console.log("KNN Classifier trained with k=3");77console.log("Test accuracy:", `${(knnAccuracy * 100).toFixed(2)}%`);78 79const probabilities = knnClassifier.predictProba(XTestKNN);80console.log("Prediction probabilities shape:", probabilities.shape);81 82// clone() returns an unfitted estimator with the same settings.83const knnFresh = knnClassifier.clone();84knnFresh.fit(XClass, yClass);85console.log(86  "Cloned classifier fitted on all rows, predicts:",87  knnFresh88    .predict(89      tensor([90        [0.5, 0.5],91        [6.5, 6.5],92      ])93    )94    .toString()95);96 97// ============================================================================98// Part 3: K-Nearest Neighbors Regression99// ============================================================================100console.log("\nPart 3: K-Nearest Neighbors Regression");101console.log("-".repeat(60));102 103const XReg = tensor([[0], [1], [2], [3], [4], [5]]);104const yReg = tensor([0, 1, 4, 9, 16, 25]);105 106const knnRegressor = new KNeighborsRegressor({ nNeighbors: 2 });107knnRegressor.fit(XReg, yReg);108 109const yPredReg = knnRegressor.predict(tensor([[2.5], [3.5]]));110console.log("Predictions for [2.5] and [3.5]:", yPredReg.toString());111 112const knnR2 = knnRegressor.score(XReg, yReg);113console.log("R² score:", knnR2.toFixed(4));114 115// ============================================================================116// Part 4: PCA (Dimensionality Reduction)117// ============================================================================118console.log("\nPart 4: PCA - Dimensionality Reduction");119console.log("-".repeat(60));120 121const XPca = tensor([122  [2.5, 2.4, 1.1],123  [0.5, 0.7, 0.3],124  [2.2, 2.9, 1.5],125  [1.9, 2.2, 0.9],126  [3.1, 3.0, 1.8],127  [2.3, 2.7, 1.2],128  [2.0, 1.6, 0.8],129  [1.0, 1.1, 0.5],130  [1.5, 1.6, 0.7],131  [1.1, 0.9, 0.4],132]);133 134const pca = new PCA({ nComponents: 2 });135pca.fit(XPca);136 137const XTransformed = pca.transform(XPca);138console.log("Original shape:", XPca.shape);139console.log("Transformed shape:", XTransformed.shape);140console.log("Explained variance ratio:", pca.explainedVarianceRatio.toString());141 142const totalVariance = Number(pca.explainedVarianceRatio.sum().item());143console.log("Total variance explained:", `${(totalVariance * 100).toFixed(2)}%`);144 145// Reconstruct data146const XReconstructed = pca.inverseTransform(XTransformed);147console.log("Reconstructed shape:", XReconstructed.shape);148 149// ============================================================================150// Part 5: Gaussian Naive Bayes151// ============================================================================152console.log("\nPart 5: Gaussian Naive Bayes");153console.log("-".repeat(60));154 155const XNB = tensor([156  [1, 2],157  [2, 3],158  [3, 4],159  [4, 5],160  [5, 6],161  [6, 7],162  [7, 8],163  [8, 9],164]);165const yNB = tensor([0, 0, 0, 0, 1, 1, 1, 1]);166 167const [XTrainNB, XTestNB, yTrainNB, yTestNB] = trainTestSplit(XNB, yNB, {168  testSize: 0.25,169  randomState: 42,170});171 172const nb = new GaussianNB();173nb.fit(XTrainNB, yTrainNB);174 175const yPredNB = nb.predict(XTestNB);176const nbAccuracy = accuracy(yTestNB, yPredNB);177 178console.log("Gaussian Naive Bayes trained");179console.log("Test accuracy:", `${(nbAccuracy * 100).toFixed(2)}%`);180 181const nbProba = nb.predictProba(XTestNB);182console.log("Prediction probabilities shape:", nbProba.shape);183 184// ============================================================================185// Summary186// ============================================================================187console.log("\nSummary");188console.log("-".repeat(60));189console.log("KMeans: unsupervised clustering of similar points");190console.log("KNN: instance-based classification and regression");191console.log("PCA: dimensionality reduction that keeps as much variance as possible");192console.log("Naive Bayes: probabilistic classifier based on Bayes' theorem");193console.log("Supervised models share fit(X, y), predict(X) and score(X, y).");194console.log("=".repeat(60));