Complete ML Pipeline

1import { mkdirSync, writeFileSync } from "node:fs";2import { loadHousingMini } from "deepbox/datasets";3import { mae, mse, r2Score } from "deepbox/metrics";4import { Ridge } from "deepbox/ml";5import { tensor } from "deepbox/ndarray";6import { Figure } from "deepbox/plot";7import { StandardScaler, trainTestSplit } from "deepbox/preprocess";8import { mean, std } from "deepbox/stats";910console.log("=".repeat(60));11console.log("Example 20: Complete Machine Learning Pipeline");12console.log("=".repeat(60));1314mkdirSync("docs/examples/12-complete-pipeline/output", { recursive: true });1516// Step 1: Load Data17console.log("\n📦 Step 1: Loading Dataset");18console.log("-".repeat(60));1920const dataset = loadHousingMini();21console.log(`✓ Loaded Housing-Mini dataset`);22console.log(`  Samples: ${dataset.data.shape[0]}`);23console.log(`  Features: ${dataset.data.shape[1]}`);2425// Step 2: Exploratory Data Analysis26console.log("\n📊 Step 2: Exploratory Data Analysis");27console.log("-".repeat(60));2829// Extract first feature for analysis30const feature_data: number[] = [];31const numFeatures = dataset.data.shape[1] || 0;32for (let i = 0; i < dataset.data.shape[0]; i++) {33  feature_data.push(Number(dataset.data.data[dataset.data.offset + i * numFeatures]));34}35const feature = tensor(feature_data);3637const meanVal = Number(mean(feature).data[0]);38const stdVal = Number(std(feature).data[0]);3940console.log(`Feature 1 Statistics:`);41console.log(`  Mean: ${meanVal.toFixed(2)}`);42console.log(`  Std:  ${stdVal.toFixed(2)}`);4344// Step 3: Data Preprocessing45console.log("\n🔄 Step 3: Data Preprocessing");46console.log("-".repeat(60));4748const [X_train, X_test, y_train, y_test] = trainTestSplit(dataset.data, dataset.target, {49  testSize: 0.2,50  randomState: 42,51  shuffle: true,52});5354console.log(`✓ Split data:`);55console.log(`  Training: ${X_train.shape[0]} samples`);56console.log(`  Testing:  ${X_test.shape[0]} samples`);5758const scaler = new StandardScaler();59scaler.fit(X_train);60const X_train_scaled = scaler.transform(X_train);61const X_test_scaled = scaler.transform(X_test);6263console.log(`✓ Scaled features using StandardScaler`);6465// Step 4: Model Training66console.log("\n🤖 Step 4: Model Training");67console.log("-".repeat(60));6869const model = new Ridge({ alpha: 1.0 });70model.fit(X_train_scaled, y_train);7172console.log(`✓ Trained Ridge Regression (α=1.0)`);7374// Step 5: Model Evaluation75console.log("\n📈 Step 5: Model Evaluation");76console.log("-".repeat(60));7778const y_pred = model.predict(X_test_scaled);7980const r2 = r2Score(y_test, y_pred);81const mseVal = mse(y_test, y_pred);82const maeVal = mae(y_test, y_pred);8384console.log(`Performance Metrics:`);85console.log(`  R² Score: ${r2.toFixed(4)}`);86console.log(`  MSE:      ${mseVal.toFixed(4)}`);87console.log(`  MAE:      ${maeVal.toFixed(4)}`);8889// Step 6: Visualization90console.log("\n🎨 Step 6: Results Visualization");91console.log("-".repeat(60));9293// Extract predictions and actual values94const y_test_array: number[] = [];95const y_pred_array: number[] = [];9697for (let i = 0; i < y_test.size; i++) {98  y_test_array.push(Number(y_test.data[y_test.offset + i]));99  y_pred_array.push(Number(y_pred.data[y_pred.offset + i]));100}101102// Create predictions vs actual plot103const fig = new Figure({ width: 640, height: 480 });104const ax = fig.addAxes();105106ax.scatter(tensor(y_test_array), tensor(y_pred_array), {107  color: "#1f77b4",108  size: 8,109});110ax.plot(tensor([0, 1, 2]), tensor([0, 1, 2]), {111  color: "#ff0000",112  linewidth: 2,113});114ax.setTitle("Predictions vs Actual Values");115ax.setXLabel("Actual");116ax.setYLabel("Predicted");117118const svg = fig.renderSVG();119writeFileSync("docs/examples/12-complete-pipeline/output/predictions.svg", svg.svg);120console.log("✓ Saved: output/predictions.svg");121122// Step 7: Summary123console.log("\n📋 Step 7: Pipeline Summary");124console.log("-".repeat(60));125126console.log(`Complete ML Pipeline Executed:`);127console.log(`  1. ✓ Data Loading (Housing-Mini dataset)`);128console.log(`  2. ✓ Exploratory Analysis`);129console.log(`  3. ✓ Train/Test Split (80/20)`);130console.log(`  4. ✓ Feature Scaling (StandardScaler)`);131console.log(`  5. ✓ Model Training (Ridge Regression)`);132console.log(`  6. ✓ Model Evaluation (R²=${r2.toFixed(3)})`);133console.log(`  7. ✓ Results Visualization`);134135console.log("\n💡 Key Takeaways:");136console.log("• Always split data before scaling to prevent data leakage");137console.log("• Feature scaling improves model performance");138console.log("• Use multiple metrics to evaluate models");139console.log("• Visualize results to understand model behavior");140console.log("• Ridge regression adds L2 regularization to prevent overfitting");141142console.log(`\n${"=".repeat(60)}`);143console.log("✅ Complete ML Pipeline Finished Successfully!");144console.log("=".repeat(60));