Complete ML Pipeline

1/**2 * Example 12: Complete ML Pipeline3 *4 * A regression workflow from start to finish on the Housing-Mini dataset:5 * load, explore, split, scale, fit a Ridge model, evaluate and plot.6 */7 8import { mkdirSync, writeFileSync } from "node:fs";9import { loadHousingMini } from "deepbox/datasets";10import { mae, mse, r2Score } from "deepbox/metrics";11import { Ridge } from "deepbox/ml";12import { tensor } from "deepbox/ndarray";13import { Figure } from "deepbox/plot";14import { StandardScaler, trainTestSplit } from "deepbox/preprocess";15import { mean, std } from "deepbox/stats";16 17console.log("=".repeat(60));18console.log("Example 12: Complete Machine Learning Pipeline");19console.log("=".repeat(60));20 21mkdirSync("docs/examples/12-complete-pipeline/output", { recursive: true });22 23// Step 1: Load data24console.log("\nStep 1: Loading the dataset");25console.log("-".repeat(60));26 27const dataset = loadHousingMini();28console.log("Loaded the Housing-Mini dataset");29console.log(`  Samples: ${dataset.data.shape[0]}`);30console.log(`  Features: ${dataset.data.shape[1]}`);31 32// Step 2: Exploratory data analysis33console.log("\nStep 2: Exploratory data analysis");34console.log("-".repeat(60));35 36// Take the first feature column. slice({}, 0) keeps every row of axis 037// and selects index 0 on axis 1.38const feature = dataset.data.slice({}, 0);39 40console.log("Feature 1 statistics:");41console.log(`  Mean: ${Number(mean(feature).item()).toFixed(2)}`);42console.log(`  Std:  ${Number(std(feature).item()).toFixed(2)}`);43 44// Step 3: Data preprocessing45console.log("\nStep 3: Data preprocessing");46console.log("-".repeat(60));47 48const [XTrain, XTest, yTrain, yTest] = trainTestSplit(dataset.data, dataset.target, {49  testSize: 0.2,50  randomState: 42,51  shuffle: true,52});53 54console.log("Split the data:");55console.log(`  Training: ${XTrain.shape[0]} samples`);56console.log(`  Testing:  ${XTest.shape[0]} samples`);57 58// Fit the scaler on the training rows only, so no test information leaks in.59const scaler = new StandardScaler();60scaler.fit(XTrain);61const XTrainScaled = scaler.transform(XTrain);62const XTestScaled = scaler.transform(XTest);63 64console.log("Scaled the features with StandardScaler");65 66// Step 4: Model training67console.log("\nStep 4: Model training");68console.log("-".repeat(60));69 70const model = new Ridge({ alpha: 1.0 });71model.fit(XTrainScaled, yTrain);72 73console.log("Trained Ridge regression (alpha = 1.0)");74 75// Step 5: Model evaluation76console.log("\nStep 5: Model evaluation");77console.log("-".repeat(60));78 79const yPred = model.predict(XTestScaled);80 81const r2 = r2Score(yTest, yPred);82const mseVal = mse(yTest, yPred);83const maeVal = mae(yTest, yPred);84 85console.log("Performance metrics:");86console.log(`  R² Score: ${r2.toFixed(4)}`);87console.log(`  MSE:      ${mseVal.toFixed(4)}`);88console.log(`  MAE:      ${maeVal.toFixed(4)}`);89 90// Step 6: Visualization91console.log("\nStep 6: Results visualization");92console.log("-".repeat(60));93 94// Predictions vs actual values. Tensors go straight into the plot.95const fig = new Figure({ width: 640, height: 480 });96const ax = fig.addAxes();97 98ax.scatter(yTest, yPred, {99  color: "#1f77b4",100  size: 8,101});102 103// The red line marks perfect predictions (predicted equals actual).104const lo = Math.min(Number(yTest.min().item()), Number(yPred.min().item()));105const hi = Math.max(Number(yTest.max().item()), Number(yPred.max().item()));106ax.plot(tensor([lo, hi]), tensor([lo, hi]), {107  color: "#ff0000",108  linewidth: 2,109});110ax.setTitle("Predictions vs Actual Values");111ax.setXLabel("Actual");112ax.setYLabel("Predicted");113 114const svg = fig.renderSVG();115writeFileSync("docs/examples/12-complete-pipeline/output/predictions.svg", svg.svg);116console.log("Saved: output/predictions.svg");117 118// Step 7: Summary119console.log("\nStep 7: Pipeline summary");120console.log("-".repeat(60));121 122console.log("Steps run:");123console.log("  1. Load data (Housing-Mini)");124console.log("  2. Exploratory analysis");125console.log("  3. Train/test split (80/20)");126console.log("  4. Feature scaling (StandardScaler)");127console.log("  5. Model training (Ridge)");128console.log(`  6. Model evaluation (R² = ${r2.toFixed(3)})`);129console.log("  7. Results visualization");130 131console.log("\nNotes:");132console.log("  Split first, then fit the scaler on the training set, to avoid data leakage.");133console.log("  Report more than one metric: R² shows fit quality, MAE is in the target's units.");134 135console.log(`\n${"=".repeat(60)}`);136console.log("Pipeline finished");137console.log("=".repeat(60));