45:"/_next/static/chunks/33yus346z0l_n.js" 46:"/_next/static/chunks/1r8ezocsukqp8.js" 47:"/_next/static/chunks/0iv56yrji_20j.js" 48:"/_next/static/chunks/0wa9m1-tphf9u.js" 49:"/_next/static/chunks/08cuqqn_f60pp.js" 4a:"/_next/static/chunks/3zdsl3f9b0--r.js" 4b:"/_next/static/chunks/2j22cxchii_xx.js" 4c:"/_next/static/chunks/3bgoyiso1rhn7.js" 4d:"/_next/static/chunks/01a-3hvnqu8wl.js" 4e:"/_next/static/chunks/3m1xjj5kw5fie.js" 4f:"/_next/static/chunks/0om3fkgfnicab.js" 50:"/_next/static/chunks/0gafyzg9j9yqj.js" 51:"/_next/static/chunks/20w4eaqlekefm.js" 52:"/_next/static/chunks/449--h5z1s6nq.js" 53:"/_next/static/chunks/3k2gyg36fnag_.js" 54:"/_next/static/chunks/18ggtgz2yf759.js" 55:"/_next/static/chunks/0hlb44ca7-4a2.js" 56:"/_next/static/chunks/253gnmj0-cj17.js" 57:"/_next/static/chunks/09nq57stwzkgf.js" 58:"/_next/static/chunks/082jo-0ajazgr.js" 59:"/_next/static/chunks/2ozcrjb_hh74t.js" 5a:"/_next/static/chunks/05i8en-x14ibi.js" 5b:"/_next/static/chunks/32izpn90gvm22.js" 5c:"/_next/static/chunks/0jbzid2v9xv3h.js" 5d:"/_next/static/chunks/0i3-eynrkvqi7.js" 5e:"/_next/static/chunks/1tn-0fivsxsv1.js" 5f:"/_next/static/chunks/34h36hyf_ggow.js" 60:"/_next/static/chunks/20329i6m_04t6.js" 61:"/_next/static/chunks/161uxd62cjny2.js" 62:"/_next/static/chunks/2k8pepwc4pxns.js" 63:I[593304,["$2b","$2c","$2d","$2e","$2f","$30","$45","$46","$47","$48","$49","$4a","$4b","$4c","$4d","$4e","$4f","$50","$51","$52","$53","$54","$55","$56","$57","$58","$59","$5a","$5b","$5c","$5d","$5e","$5f","$60","$61","$62"],"OwnedHydration"] 64:T46c,In 1995, Yoav Freund and Robert Schapire introduced **AdaBoost** (Adaptive Boosting)—an algorithm that would fundamentally transform machine learning and earn its creators the prestigious Gödel Prize in 2003. AdaBoost provided the first practical answer to a profound theoretical question: *Can weak learners be combined to create arbitrarily accurate predictors?* The answer, as AdaBoost demonstrated, is a resounding **yes**. By intelligently focusing on misclassified examples and combining weak classifiers through weighted voting, AdaBoost achieved classification accuracies that seemed almost magical at the time. Viola and Jones's face detection system—which used AdaBoost at its core—remains one of the most successful applications of machine learning in computer vision history. This page provides an exhaustive treatment of the AdaBoost algorithm: its historical context, theoretical foundations, complete mathematical derivation, step-by-step procedure, and practical implementation considerations. By the end, you will not merely understand AdaBoost—you will be able to derive it, implement it, and extend it.65:T557,The story of AdaBoost begins with a fundamental question in computational learning theory. In 1988, Michael Kearns posed what became known as the **boosting problem**: *Is it possible to combine multiple "weak" learning algorithms—each performing only slightly better than random guessing—into a single "strong" learner with arbitrarily high accuracy?* This question might seem strange at first. Why would we want to combine bad classifiers? The intuition comes from an analogy with human decision-making. Consider a panel of experts, where each expert is only slightly better than chance at predicting outcomes. Individually, their predictions are nearly useless. But if their errors are *independent* and we aggregate their votes intelligently, the combined judgment can be remarkably accurate—this is the **wisdom of crowds** phenomenon. **The Weak Learning Hypothesis:** A **weak learner** is defined formally as an algorithm that, for any distribution over training examples, produces a classifier with error rate at most $\frac{1}{2} - \gamma$ for some $\gamma > 0$. In other words: $$\varepsilon = P(h(x) eq y) \leq \frac{1}{2} - \gamma$$ The parameter $\gamma$ represents the "edge" over random guessing. Even if $\gamma = 0.01$ (the classifier is 51% accurate), boosting theory promises we can amplify this to near-perfect classification.66:T6f1,Before diving into mathematics, let's build a deep intuitive understanding of how AdaBoost works. The algorithm's genius lies in three interconnected ideas: **Idea 1: Focus on Mistakes** Consider training a classifier on a dataset. After training, some examples are correctly classified, others are misclassified. A natural question arises: *What should the next classifier focus on?* AdaBoost's answer is elegant: **increase the importance of misclassified examples**. By reweighting the training data so that mistakes matter more, the next weak learner is forced to pay attention to the examples the previous learner got wrong. This creates a sequence of classifiers, each compensating for its predecessor's failures. **Idea 2: Combine Through Weighted Voting** Once we've trained multiple weak learners, how do we combine their predictions? AdaBoost uses **weighted majority voting**, where classifiers with lower error rates receive higher voting weights. This is intuitive: we trust accurate classifiers more than inaccurate ones. The key insight is that the voting weights are determined during training, based on each classifier's performance on the weighted training data. A classifier with 49% error (barely better than chance) gets a small weight; a classifier with 5% error gets a large weight. **Idea 3: Adaptivity is Key** The "adaptive" in AdaBoost refers to both: 1. **Adaptive data weighting**: The distribution over examples changes based on which examples are misclassified 2. **Adaptive classifier weighting**: Each classifier's vote is weighted by its demonstrated accuracy This dual adaptivity creates a powerful feedback loop: weak learners specialize on different subsets of the data, and their contributions are calibrated to their reliability.67:T65d,Now let's formalize AdaBoost mathematically. We work with a **binary classification** setting, which is the canonical case for AdaBoost. **Problem Setup:** Given: - Training set: $\{(x_1, y_1), (x_2, y_2), \ldots, (x_n, y_n)\}$ - Labels: $y_i \in \{-1, +1\}$ (bipolar encoding) - Weak learner: Returns classifier $h_t: \mathcal{X} \to \{-1, +1\}$ for iteration $t$ - Number of boosting rounds: $T$ **Key Variables:** - $D_t(i)$: Weight of example $i$ at round $t$ (satisfies $\sum_{i=1}^n D_t(i) = 1$) - $h_t$: Weak classifier returned at round $t$ - $\varepsilon_t$: Weighted error of $h_t$ ($\varepsilon_t = \sum_{i: h_t(x_i) eq y_i} D_t(i)$) - $\alpha_t$: Voting weight for classifier $h_t$ **The Algorithm:** **Initialization (t = 0):** Initialize example weights uniformly: $$D_1(i) = \frac{1}{n} \quad \text{for } i = 1, 2, \ldots, n$$ **For t = 1, 2, ..., T:** 1. **Train weak learner** using distribution $D_t$, obtaining $h_t$ 2. **Compute weighted error:** $$\varepsilon_t = \sum_{i=1}^{n} D_t(i) \cdot \mathbb{1}[h_t(x_i) eq y_i] = \sum_{i: h_t(x_i) eq y_i} D_t(i)$$ 3. **Compute classifier weight:** $$\alpha_t = \frac{1}{2} \ln\left(\frac{1 - \varepsilon_t}{\varepsilon_t}\right)$$ 4. **Update example weights:** $$D_{t+1}(i) = \frac{D_t(i) \cdot \exp(-\alpha_t \cdot y_i \cdot h_t(x_i))}{Z_t}$$ where $Z_t$ is a normalization constant: $$Z_t = \sum_{i=1}^{n} D_t(i) \cdot \exp(-\alpha_t \cdot y_i \cdot h_t(x_i))$$ **Final Classifier:** The strong classifier is a weighted majority vote: $$H(x) = \text{sign}\left(\sum_{t=1}^{T} \alpha_t \cdot h_t(x)\right)$$68:T487,**Understanding the Formulas:** **Why this formula for $\alpha_t$?** The classifier weight $\alpha_t = \frac{1}{2}\ln\left(\frac{1-\varepsilon_t}{\varepsilon_t}\right)$ has deep significance: - When $\varepsilon_t = 0.5$ (random guessing): $\alpha_t = 0$ (zero weight) - When $\varepsilon_t \to 0$ (perfect classification): $\alpha_t \to \infty$ (very high weight) - When $\varepsilon_t \to 0.5$: $\alpha_t \to 0$ (approaches zero weight) This formula arises naturally from minimizing the exponential loss, as we'll see in a later page. **Why this update for $D_{t+1}$?** The weight update has an elegant interpretation: - If $h_t(x_i) = y_i$ (correct): $y_i \cdot h_t(x_i) = +1$, so $D_{t+1}(i) \propto D_t(i) \cdot e^{-\alpha_t}$ (weight **decreases**) - If $h_t(x_i) eq y_i$ (incorrect): $y_i \cdot h_t(x_i) = -1$, so $D_{t+1}(i) \propto D_t(i) \cdot e^{+\alpha_t}$ (weight **increases**) Since $\alpha_t > 0$ when $\varepsilon_t < 0.5$, correctly classified examples lose weight while misclassified examples gain weight. This is the adaptive reweighting that forces subsequent learners to focus on hard examples.69:T617,We'll use **decision stumps** (single-feature thresholds) as weak learners. **Initialization:** $D_1(i) = \frac{1}{6}$ for all $i = 1, \ldots, 6$ **Round 1:** The weak learner finds the best decision stump. Suppose it selects: $h_1(x) = +1$ if $x_1 < 2.5$, else $-1$. Predictions: (1:+1, 2:+1, 3:-1, 4:+1, 5:-1, 6:-1) Actual: (+1, +1, -1, +1, -1, -1) This classifies **example 5** incorrectly (predicts -1, actual is -1... wait, let me recalculate) Actually: h₁ predictions for each example: - x₁=1 < 2.5: h₁=+1 (example 1, label +1) ✓ - x₁=2 < 2.5: h₁=+1 (example 2, label +1) ✓ - x₁=3 ≥ 2.5: h₁=-1 (example 3, label -1) ✓ - x₁=1 < 2.5: h₁=+1 (example 4, label +1) ✓ - x₁=2 < 2.5: h₁=+1 (example 5, label -1) ✗ - x₁=3 ≥ 2.5: h₁=-1 (example 6, label -1) ✓ Weighted error: $\varepsilon_1 = D_1(5) = \frac{1}{6} \approx 0.167$ Classifier weight: $\alpha_1 = \frac{1}{2}\ln\left(\frac{1 - 0.167}{0.167}\right) = \frac{1}{2}\ln(5) \approx 0.805$ **Weight Update for Round 2:** For correctly classified (i ∈ {1,2,3,4,6}): $D_2(i) \propto \frac{1}{6} \cdot e^{-0.805} \approx \frac{1}{6} \cdot 0.447 = 0.0745$ For incorrectly classified (i = 5): $D_2(5) \propto \frac{1}{6} \cdot e^{+0.805} \approx \frac{1}{6} \cdot 2.237 = 0.373$ After normalization (Z₁ = 5 × 0.0745 + 0.373 = 0.746): - D₂(1) = D₂(2) = D₂(3) = D₂(4) = D₂(6) ≈ 0.0745/0.746 = 0.10 - D₂(5) ≈ 0.373/0.746 = 0.50 Notice how example 5's weight jumped from 0.167 to 0.50—it now dominates the distribution!6a:T942,import numpy as np class DecisionStump: """ A decision stump weak learner for AdaBoost. Finds optimal threshold on best feature to minimize weighted error. """ def __init__(self): self.feature_idx = None self.threshold = None self.polarity = 1 # 1 or -1 def fit(self, X: np.ndarray, y: np.ndarray, weights: np.ndarray): """ Find optimal decision stump for weighted data. Args: X: Feature matrix (n_samples, n_features) y: Labels in {-1, +1} weights: Sample weights (must sum to 1) """ n_samples, n_features = X.shape min_error = float('inf') for feature_idx in range(n_features): # Get unique thresholds (midpoints between sorted values) feature_values = X[:, feature_idx] sorted_values = np.sort(np.unique(feature_values)) if len(sorted_values) == 1: thresholds = sorted_values else: thresholds = (sorted_values[:-1] + sorted_values[1:]) / 2 for threshold in thresholds: for polarity in [1, -1]: # Make predictions predictions = np.ones(n_samples) if polarity == 1: predictions[feature_values < threshold] = -1 else: predictions[feature_values >= threshold] = -1 # Compute weighted error error = np.sum(weights[predictions != y]) if error < min_error: min_error = error self.feature_idx = feature_idx self.threshold = threshold self.polarity = polarity return min_error def predict(self, X: np.ndarray) -> np.ndarray: """Return predictions in {-1, +1}.""" n_samples = X.shape[0] predictions = np.ones(n_samples) feature_values = X[:, self.feature_idx] if self.polarity == 1: predictions[feature_values < self.threshold] = -1 else: predictions[feature_values >= self.threshold] = -1 return predictions6b:T1260,import numpy as np from typing import List, Tuple class AdaBoost: """ AdaBoost (Adaptive Boosting) classifier. Combines weak learners through weighted majority voting, with weights determined by each learner's accuracy on adaptively reweighted training data. """ def __init__(self, n_estimators: int = 50, min_error: float = 1e-10): """ Args: n_estimators: Number of boosting rounds (T) min_error: Minimum error to prevent numerical issues """ self.n_estimators = n_estimators self.min_error = min_error self.estimators: List[DecisionStump] = [] self.alphas: List[float] = [] def fit(self, X: np.ndarray, y: np.ndarray) -> 'AdaBoost': """ Fit the AdaBoost classifier. Args: X: Training features (n_samples, n_features) y: Training labels in {-1, +1} Returns: self """ n_samples = X.shape[0] # Validate labels assert set(np.unique(y)) <= {-1, 1}, "Labels must be in {-1, +1}" # Initialize uniform weights weights = np.ones(n_samples) / n_samples for t in range(self.n_estimators): # Train weak learner on weighted data stump = DecisionStump() error = stump.fit(X, y, weights) # Clip error to avoid numerical issues error = np.clip(error, self.min_error, 1 - self.min_error) # Check weak learning condition if error >= 0.5: print(f"Round {t}: Error {error:.4f} >= 0.5, stopping early") break # Compute classifier weight alpha = 0.5 * np.log((1 - error) / error) # Get predictions predictions = stump.predict(X) # Update weights # weights *= exp(-alpha * y * predictions) weights *= np.exp(-alpha * y * predictions) # Normalize weights weights /= np.sum(weights) # Store classifier and weight self.estimators.append(stump) self.alphas.append(alpha) # Progress logging train_error = np.mean(self.predict(X) != y) print(f"Round {t+1}: ε={error:.4f}, α={alpha:.4f}, " f"train_error={train_error:.4f}") return self def predict_raw(self, X: np.ndarray) -> np.ndarray: """ Compute raw weighted vote (before sign). Returns sum of alpha_t * h_t(x) for all t. """ n_samples = X.shape[0] raw_scores = np.zeros(n_samples) for stump, alpha in zip(self.estimators, self.alphas): raw_scores += alpha * stump.predict(X) return raw_scores def predict(self, X: np.ndarray) -> np.ndarray: """ Predict class labels for samples. Returns labels in {-1, +1}. """ return np.sign(self.predict_raw(X)) def predict_proba(self, X: np.ndarray) -> np.ndarray: """ Predict class probabilities using logistic transform. Returns (n_samples, 2) array with P(y=-1) and P(y=+1). """ raw_scores = self.predict_raw(X) # Use logistic function to convert to probabilities proba_positive = 1 / (1 + np.exp(-2 * raw_scores)) proba_negative = 1 - proba_positive return np.column_stack([proba_negative, proba_positive]) def staged_predict(self, X: np.ndarray): """ Yield predictions after each boosting round. Useful for analyzing training dynamics. """ n_samples = X.shape[0] raw_scores = np.zeros(n_samples) for stump, alpha in zip(self.estimators, self.alphas): raw_scores += alpha * stump.predict(X) yield np.sign(raw_scores) # Example usage if __name__ == "__main__": # Generate sample data np.random.seed(42) n_samples = 200 # Two overlapping Gaussian clusters X_pos = np.random.randn(n_samples // 2, 2) + [1, 1] X_neg = np.random.randn(n_samples // 2, 2) + [-1, -1] X = np.vstack([X_pos, X_neg]) y = np.array([1] * (n_samples // 2) + [-1] * (n_samples // 2)) # Train AdaBoost clf = AdaBoost(n_estimators=20) clf.fit(X, y) # Evaluate predictions = clf.predict(X) accuracy = np.mean(predictions == y) print(f" Final training accuracy: {accuracy:.4f}")6c:T46c,One of AdaBoost's most striking properties is its ability to drive training error to zero remarkably quickly—often within just a few dozen rounds. **Training Error Bound:** Freund and Schapire proved that the training error of the final classifier satisfies: $$\text{TrainErr}(H) \leq \prod_{t=1}^{T} Z_t = \prod_{t=1}^{T} 2\sqrt{\varepsilon_t(1 - \varepsilon_t)}$$ If each weak learner has edge $\gamma_t = \frac{1}{2} - \varepsilon_t$, this simplifies to: $$\text{TrainErr}(H) \leq \prod_{t=1}^{T} \sqrt{1 - 4\gamma_t^2} \leq \exp\left(-2\sum_{t=1}^{T} \gamma_t^2\right)$$ If $\gamma_t \geq \gamma > 0$ for all $t$, the training error decays exponentially: $$\text{TrainErr}(H) \leq e^{-2\gamma^2 T}$$ **Practical Implications:** This bound is extraordinarily powerful. Even with weak learners that have only 5% edge (55% accuracy), after 100 rounds: $$\text{TrainErr} \leq e^{-2 \times 0.05^2 \times 100} = e^{-0.5} \approx 0.606$$ After 500 rounds: $$\text{TrainErr} \leq e^{-2.5} \approx 0.082$$ After 2000 rounds: $$\text{TrainErr} \leq e^{-10} \approx 0.000045$$ The training error plummets exponentially!44:["$","$L63",null,{"owner":"anonymous","state":{"mutations":[],"queries":[{"dehydratedAt":1792320023902,"state":{"data":{"json":{"id":"seed_150db48de7956ac135d299886c203e81","slug":"adaboost","category":"Machine Learning","topic":"AdaBoost","title":"AdaBoost: Adaptive Boosting","description":"Master the AdaBoost algorithm—the seminal boosting method that revolutionized ensemble learning. Understand adaptive weight updates, weighted voting, theoretical error bounds, and the deep connection to exponential loss minimization.","difficulty":"ADVANCED","estimatedMinutes":90,"pages":[{"id":"seed_9f9509db19eb4a045548fb42acc478a4","title":"Adaptive Boosting Algorithm","order":0,"content":[{"id":"opening-hook","title":"The Birth of Practical Boosting","content":[{"data":{"content":"$64"},"type":"text"},{"data":{"icon":"Target","title":"What You Will Learn","content":"By completing this page, you will: (1) Understand the historical development and motivation for AdaBoost; (2) Master the complete mathematical formulation of the algorithm; (3) Trace through the algorithm step-by-step with concrete examples; (4) Understand the role of weak learners and their requirements; (5) Be prepared to implement AdaBoost from scratch in any programming language.","variant":"info"},"type":"callout"}]},{"id":"historical-context","title":"Historical Context and Motivation","content":[{"data":{"content":"$65"},"type":"text"},{"data":{"icon":"Info","title":"The Weak Learning Condition","content":"The weak learning condition requires only that the learner perform better than random guessing. For binary classification, this means accuracy > 50%. This seemingly modest requirement has profound implications: if *any* such learner exists for a problem, boosting can achieve arbitrarily high accuracy.","variant":"note"},"type":"callout"},{"data":{"content":"**From Theory to Practice:**\n\nIn 1990, Robert Schapire proved that boosting is indeed possible, demonstrating the theoretical equivalence of weak and strong learnability. However, his initial algorithm was complex and impractical. The breakthrough came in 1995 when Freund and Schapire introduced AdaBoost—an algorithm that was not only theoretically sound but remarkably simple and practical.\n\nAdaBoost's key insight was to maintain a **distribution over training examples** that adapts based on classification errors. Examples that are misclassified receive higher weight, forcing subsequent weak learners to focus on \"hard\" cases. This adaptive reweighting—the \"Ada\" in AdaBoost—is what gives the algorithm its power.\n\n**Timeline of Key Developments:**"},"type":"text"},{"data":{"rows":[["1988","Kearns poses boosting problem","Foundational question formulated"],["1990","Schapire's first boosting algorithm","Proves boosting is possible"],["1995","Freund & Schapire introduce AdaBoost","Practical, efficient boosting"],["1996","Theoretical analysis tightened","Margin-based generalization bounds"],["1997","AdaBoost.M1, M2 for multiclass","Extensions to multiple classes"],["2001","Viola-Jones face detection","Landmark real-world application"],["2003","Gödel Prize to Freund & Schapire","Recognition of theoretical importance"]],"headers":["Year","Development","Significance"]},"type":"table"}],"titleType":"number","titleNumber":1},{"id":"core-intuition","title":"The Core Intuition","content":[{"data":{"content":"$66"},"type":"text"},{"data":{"left":[{"data":{"items":["**Initialize** uniform weights over examples","**Train** weak learner on weighted data","**Compute** classifier weight from error","**Update** example weights (increase for mistakes)","**Repeat** for T iterations","**Combine** via weighted majority vote"],"title":{"icon":"Target","text":"The AdaBoost Strategy","type":"icon","color":"primary","variant":"outline-circle"},"variant":"accent"},"type":"list"}],"right":[{"data":{"items":["Mistakes become **high-priority** for next round","Weak learners **specialize** on hard cases","Accurate classifiers get **more voting power**","Combined predictions **average out errors**","Theoretical guarantees on **training error**","Empirically strong **generalization**"],"title":{"icon":"Lightbulb","text":"Why It Works","type":"icon","color":"accent","variant":"outline-circle"},"variant":"accent"},"type":"list"}]},"type":"two-column"},{"data":{"icon":"Lightbulb","title":"A Helpful Analogy","content":"Imagine a student repeatedly taking practice exams. After each exam, the student studies the questions they got wrong more intensively. Over time, they develop competence across all types of questions. AdaBoost works similarly: each weak learner is like an exam, and the weight updates ensure comprehensive coverage of the problem space.","variant":"tip"},"type":"callout"}],"titleType":"number","titleNumber":2},{"id":"mathematical-formulation","title":"Mathematical Formulation","content":[{"data":{"content":"$67"},"type":"text"},{"data":{"icon":"AlertCircle","title":"Critical Requirement","content":"The algorithm requires ε_t < 0.5 (better than random guessing) at each iteration. If a weak learner performs at or worse than chance, the algorithm fails. In practice, this is rarely an issue with reasonable weak learners like decision stumps.","variant":"warning"},"type":"callout"},{"data":{"content":"$68"},"type":"text"},{"data":{"rows":[["0.001","3.45","Excellent classifier, very high weight"],["0.05","1.47","Good classifier, high weight"],["0.10","1.10","Decent classifier, moderate-high weight"],["0.20","0.69","Moderate classifier, moderate weight"],["0.30","0.42","Weak classifier, low weight"],["0.40","0.20","Very weak classifier, very low weight"],["0.49","0.02","Barely better than random, near-zero weight"],["0.50","0.00","Random guessing, zero weight"]],"caption":"Effect of Error Rate on Classifier Weight","headers":["Error Rate (ε)","α Value","Interpretation"]},"type":"table"}],"titleType":"number","titleNumber":3},{"id":"step-by-step","title":"Step-by-Step Walkthrough","content":[{"data":{"content":"Let's trace through AdaBoost with a concrete example. Consider a simple toy dataset:"},"type":"text"},{"data":{"rows":[["1","1","1","+1"],["2","2","1","+1"],["3","3","1","-1"],["4","1","2","+1"],["5","2","2","-1"],["6","3","2","-1"]],"headers":["Example","x₁","x₂","y (Label)"]},"type":"table"},{"data":{"content":"$69"},"type":"text"},{"data":{"icon":"Lightbulb","title":"The Focusing Effect","content":"After just one round, the misclassified example (5) now has 5× the weight of correctly classified examples. The next weak learner must classify it correctly or suffer a high weighted error. This is AdaBoost's mechanism for ensuring comprehensive coverage.","variant":"tip"},"type":"callout"},{"data":{"content":"**Round 2:**\n\nWith the new distribution heavily weighted toward example 5, the weak learner finds a stump that correctly classifies it.\n\nSuppose it selects: \$h_2(x) = -1\$ if \$x_2 \\geq 1.5\$, else \$+1\$.\n\nPredictions: (1:+1, 2:+1, 3:+1, 4:-1, 5:-1, 6:-1)\n\nErrors on examples 3 (pred +1, actual -1) and 4 (pred -1, actual +1).\n\nWeighted error: \$\\varepsilon_2 = D_2(3) + D_2(4) = 0.10 + 0.10 = 0.20\$\n\nClassifier weight: \$\\alpha_2 = \\frac{1}{2}\\ln\\left(\\frac{0.80}{0.20}\\right) = \\frac{1}{2}\\ln(4) \\approx 0.693\$\n\n**Round 3 and Beyond:**\n\nThe process continues. Example 5 is now correctly classified, so its weight decreases. Examples 3 and 4 are now the \"hard\" cases, so their weights increase. Each round focuses on different problem areas.\n\n**Final Classifier (after T rounds):**\n\n$$H(x) = \\text{sign}(0.805 \\cdot h_1(x) + 0.693 \\cdot h_2(x) + \\alpha_3 \\cdot h_3(x) + \\ldots)$$\n\nThe final prediction is a weighted vote of all weak learners, with weights proportional to their accuracy."},"type":"text"}],"titleType":"number","titleNumber":4},{"id":"weak-learners","title":"Weak Learners in Practice","content":[{"data":{"content":"The choice of weak learner is crucial to AdaBoost's success. While any learner satisfying the weak learning condition works theoretically, practical considerations favor certain choices.\n\n**Decision Stumps: The Default Choice**\n\nA **decision stump** is a decision tree with a single split—the simplest possible tree. It tests one feature against one threshold and assigns different labels to each side.\n\nFor continuous feature \$x_j\$ and threshold \$\\theta\$:\n$$h(x) = \\begin{cases} +1 & \\text{if } x_j < \\theta \\\\ -1 & \\text{otherwise} \\end{cases}$$\n\nDecision stumps are the most common weak learner for AdaBoost because:\n\n- **Simplicity**: O(n·d) to find optimal stump (n examples, d features)\n- **Interpretability**: Each stump is a single rule\n- **Weak but not too weak**: Typically achieves 30-45% error on weighted data\n- **Fast training**: Critical when training hundreds of weak learners"},"type":"text"},{"data":{"rows":[["Decision stump","O(n·d·log n)","30-45%","Fast, interpretable","Very weak individually"],["Depth-2 tree","O(n·d²·log n)","20-35%","More expressive","Slower, less interpretable"],["Depth-3 tree","O(n·d³·log n)","15-30%","Good balance","Risk of overfitting"],["Linear classifier","O(n·d)","20-40%","Fast, smooth boundaries","May be too strong"],["Naive Bayes","O(n·d)","25-40%","Handles mixed features","Independence assumption"]],"caption":"Comparison of Weak Learners for AdaBoost","headers":["Weak Learner","Training Time","Typical Error","Pros","Cons"]},"type":"table"},{"data":{"content":"**The Goldilocks Principle:**\n\nThe weak learner should be **weak enough** that it doesn't overfit but **strong enough** to exceed random chance. If weak learners are too strong:\n- They may memorize training data\n- Less room for boosting to help\n- Risk of overfitting ensemble\n\nIf weak learners are too weak:\n- Need many more boosting rounds\n- May not converge\n- Risk of underfitting\n\n**Decision stumps hit the sweet spot for most problems.**\n\n**Handling the Weighted Distribution:**\n\nWeak learners must handle the weighted distribution \$D_t\$. There are two approaches:\n\n1. **Weight-aware training**: The learner directly uses weights in its objective\n - For decision stumps: weighted misclassification rate\n - For linear models: weighted loss function\n\n2. **Resampling**: Draw a bootstrap sample according to \$D_t\$, train on unweighted sample\n - Creates an effective weighted distribution\n - Useful for learners that don't support weights natively\n - Introduces additional randomness"},"type":"text"},{"data":{"fileName":"decision_stump.py","languages":[{"code":"$6a","label":"Python","value":"python"}],"showLineNumbers":true},"type":"code"}],"titleType":"number","titleNumber":5},{"id":"implementation","title":"Complete AdaBoost Implementation","content":[{"data":{"content":"Armed with our understanding of the algorithm and weak learners, here is a complete, production-quality implementation of AdaBoost:"},"type":"text"},{"data":{"fileName":"adaboost.py","languages":[{"code":"$6b","label":"Python","value":"python"}],"showLineNumbers":true},"type":"code"},{"data":{"icon":"Info","title":"Implementation Details","content":"The implementation includes error clipping to prevent numerical overflow in the α calculation, staged prediction for analyzing training dynamics, and probability estimation via logistic transform. These are essential for production use but often omitted in textbook descriptions.","variant":"note"},"type":"callout"}],"titleType":"number","titleNumber":6},{"id":"convergence","title":"Convergence Behavior","content":[{"data":{"content":"$6c"},"type":"text"},{"data":{"icon":"CheckCircle","title":"The Power of Boosting","content":"AdaBoost can achieve zero training error with enough rounds, even when using the simplest weak learners. This seemed almost magical when first discovered—how can combining many bad classifiers produce a perfect one? The answer lies in the adaptive reweighting, which ensures each weak learner contributes something new.","variant":"success"},"type":"callout"},{"data":{"content":"**What About Generalization?**\n\nTraining error going to zero sounds like a recipe for overfitting. Remarkably, AdaBoost often does **not** overfit, even when training error reaches zero. The **margin theory** explains this:\n\nAdaBoost doesn't just minimize errors—it increases **margins** (the confidence of predictions). Even after training error hits zero, subsequent rounds increase the margin on hard examples, improving generalization.\n\nWe'll explore margin theory and its implications in a later page. For now, understand that AdaBoost's behavior is more sophisticated than the training error bound suggests.\n\n**Typical Training Dynamics:**"},"type":"text"},{"data":{"rows":[["1-5","20-40%","Initial rapid descent"],["5-20","5-20%","Continued improvement"],["20-50","0-5%","Approaching convergence"],["50-100","~0%","Often at or near zero"],["100+","0%","Margin improvement phase"]],"headers":["Round","Training Error","Observations"]},"type":"table"}],"titleType":"number","titleNumber":7},{"id":"summary","title":"Summary and Key Takeaways","content":[{"data":{"content":"We've covered the AdaBoost algorithm in exhaustive detail. Let's consolidate the essential knowledge:"},"type":"text"},{"data":{"type":"ordered","items":["**Adaptive Boosting**: AdaBoost adaptively reweights training examples, focusing on misclassified samples in subsequent rounds.","**Weak Learning Condition**: Each weak learner must achieve error < 0.5 on the weighted distribution—barely better than random guessing suffices.","**Weight Update Formula**: D_{t+1}(i) ∝ D_t(i) · exp(-α_t · y_i · h_t(x_i)) increases weights for mistakes, decreases for correct predictions.","**Classifier Weighting**: α_t = ½ ln((1-ε_t)/ε_t) gives higher voting power to more accurate classifiers.","**Final Classifier**: H(x) = sign(Σ α_t · h_t(x)) is a weighted majority vote of all weak learners.","**Exponential Convergence**: Training error decreases exponentially with the number of boosting rounds.","**Decision Stumps**: Single-split trees are the most common weak learner—simple, fast, and effective."],"title":{"icon":"CheckCircle","text":"Key Concepts to Remember","type":"icon","color":"primary","variant":"outline-circle"},"variant":"accent"},"type":"list"},{"data":{"icon":"Zap","title":"What's Next","content":"The next page dives deep into weight updates—the mathematical heart of AdaBoost. We'll derive the update formula from first principles, understand its connection to exponential loss, and see why this specific update schedule leads to optimal performance.","variant":"tip"},"type":"callout"},{"data":{"icon":"CheckCircle","title":"Page Complete","content":"You now have a comprehensive understanding of the AdaBoost algorithm: its historical development, core intuition, complete mathematical formulation, and practical implementation. This foundational knowledge prepares you for the deeper topics ahead: weight updates, error bounds, and the connection to exponential loss.","variant":"success"},"type":"callout"}],"titleType":"number","titleNumber":8}],"isLocked":false},{"id":"seed_b0ef2a1154c2561572a94854fcee8d64","title":"Weight Updates","order":1,"content":null,"isLocked":true},{"id":"seed_c7c9c97ada179ecb486c200accc4fc6f","title":"Weighted Voting","order":2,"content":null,"isLocked":true},{"id":"seed_348190b4ed209df5ea6f490204fe09a9","title":"Training Error Bounds","order":3,"content":null,"isLocked":true},{"id":"seed_24e976d10f8f97156a4382c0db3dca8f","title":"Exponential Loss Connection","order":4,"content":null,"isLocked":true}]}},"dataUpdateCount":1,"dataUpdatedAt":1792320023896,"error":null,"errorUpdateCount":0,"errorUpdatedAt":0,"fetchFailureCount":0,"fetchFailureReason":null,"fetchMeta":null,"isInvalidated":false,"status":"success","fetchStatus":"idle"},"queryKey":[["learningContent","getModule"],{"input":{"slug":"adaboost"},"type":"query"}],"queryHash":"[[\"learningContent\",\"getModule\"],{\"input\":{\"slug\":\"adaboost\"},\"type\":\"query\"}]"},{"dehydratedAt":1792320023903,"state":{"data":{"json":null},"dataUpdateCount":1,"dataUpdatedAt":1792320023901,"error":null,"errorUpdateCount":0,"errorUpdatedAt":0,"fetchFailureCount":0,"fetchFailureReason":null,"fetchMeta":null,"isInvalidated":false,"status":"success","fetchStatus":"idle"},"queryKey":[["learningContent","getProgress"],{"input":{"moduleId":"seed_150db48de7956ac135d299886c203e81"},"type":"query"}],"queryHash":"[[\"learningContent\",\"getProgress\"],{\"input\":{\"moduleId\":\"seed_150db48de7956ac135d299886c203e81\"},\"type\":\"query\"}]"},{"dehydratedAt":1792320023903,"state":{"data":{"json":null},"dataUpdateCount":1,"dataUpdatedAt":1792320023902,"error":null,"errorUpdateCount":0,"errorUpdatedAt":0,"fetchFailureCount":0,"fetchFailureReason":null,"fetchMeta":null,"isInvalidated":false,"status":"success","fetchStatus":"idle"},"queryKey":[["feedback","get"],{"input":{"contentType":"LEARNING_PAGE","contentId":"seed_b0ef2a1154c2561572a94854fcee8d64"},"type":"query"}],"queryHash":"[[\"feedback\",\"get\"],{\"input\":{\"contentId\":\"seed_b0ef2a1154c2561572a94854fcee8d64\",\"contentType\":\"LEARNING_PAGE\"},\"type\":\"query\"}]"},{"dehydratedAt":1792320023903,"state":{"data":{"json":null},"dataUpdateCount":1,"dataUpdatedAt":1792320023902,"error":null,"errorUpdateCount":0,"errorUpdatedAt":0,"fetchFailureCount":0,"fetchFailureReason":null,"fetchMeta":null,"isInvalidated":false,"status":"success","fetchStatus":"idle"},"queryKey":[["feedback","getUserLike"],{"input":{"contentType":"LEARNING_PAGE","contentId":"seed_b0ef2a1154c2561572a94854fcee8d64"},"type":"query"}],"queryHash":"[[\"feedback\",\"getUserLike\"],{\"input\":{\"contentId\":\"seed_b0ef2a1154c2561572a94854fcee8d64\",\"contentType\":\"LEARNING_PAGE\"},\"type\":\"query\"}]"},{"dehydratedAt":1792320023903,"state":{"data":{"json":[]},"dataUpdateCount":1,"dataUpdatedAt":1792320023902,"error":null,"errorUpdateCount":0,"errorUpdatedAt":0,"fetchFailureCount":0,"fetchFailureReason":null,"fetchMeta":null,"isInvalidated":false,"status":"success","fetchStatus":"idle"},"queryKey":[["highlights","getHighlights"],{"input":{"contentType":"LEARNING_PAGE","contentId":"seed_b0ef2a1154c2561572a94854fcee8d64"},"type":"query"}],"queryHash":"[[\"highlights\",\"getHighlights\"],{\"input\":{\"contentId\":\"seed_b0ef2a1154c2561572a94854fcee8d64\",\"contentType\":\"LEARNING_PAGE\"},\"type\":\"query\"}]"},{"dehydratedAt":1792320023903,"state":{"data":"$undefined","dataUpdateCount":0,"dataUpdatedAt":0,"error":null,"errorUpdateCount":0,"errorUpdatedAt":0,"fetchFailureCount":0,"fetchFailureReason":null,"fetchMeta":null,"isInvalidated":false,"status":"pending","fetchStatus":"fetching"},"queryKey":[["highlights","getPopularHighlights"],{"input":{"contentType":"LEARNING_PAGE","contentId":"seed_b0ef2a1154c2561572a94854fcee8d64","minCount":2},"type":"query"}],"queryHash":"[[\"highlights\",\"getPopularHighlights\"],{\"input\":{\"contentId\":\"seed_b0ef2a1154c2561572a94854fcee8d64\",\"contentType\":\"LEARNING_PAGE\",\"minCount\":2},\"type\":\"query\"}]","promise":"$@6d"}]},"children":"$L6e"}]