Artifact of #18

Code

# Breiman (2001) Table 2, row 'Diabetes', column 'Adaboost' (paper: 26.6 %; same row: Selection 24.2, Forest-RI single input 24.3, one tree 33.1).
# Pima Indians diabetes, 768 rows, 8 inputs, original file with zeros left as they are. Random 10% hold-out (76 cases), 100 repetitions averaged.
# Adaboost with 50 fully grown CART trees, Breiman's resampling form: each tree is grown on a bootstrap sample drawn with the current weights,
# the weighted error eps is measured on the whole training set, weights of misclassified cases are multiplied by (1-eps)/eps and renormalised,
# and if eps = 0 or eps >= 1/2 the weights are reset to uniform. The vote of a tree is log((1-eps)/eps). One average = one seed.
import sys, urllib.request
import numpy as np
from sklearn.tree import DecisionTreeClassifier
URL = "https://raw.githubusercontent.com/jbrownlee/Datasets/master/pima-indians-diabetes.data.csv"
raw = urllib.request.urlopen(URL, timeout=60).read().decode()
rows = [r.split(",") for r in raw.strip().splitlines()]
X = np.array([[float(v) for v in r[:8]] for r in rows]); y = np.array([int(r[8]) for r in rows])
n, M = X.shape
T = 50
def adaboost(Xtr, ytr, Xte, rng):
    m = len(ytr); w = np.full(m, 1.0 / m); score = np.zeros(len(Xte))
    for _ in range(T):
        idx = rng.choice(m, size=m, p=w)
        tree = DecisionTreeClassifier(random_state=int(rng.integers(2**31 - 1))).fit(Xtr[idx], ytr[idx])
        wrong = tree.predict(Xtr) != ytr
        eps = w[wrong].sum()
        if eps == 0 or eps >= 0.5:
            w = np.full(m, 1.0 / m); continue
        beta = (1 - eps) / eps
        w = np.where(wrong, w * beta, w); w /= w.sum()
        score += np.log(beta) * np.where(tree.predict(Xte) == 1, 1.0, -1.0)
    return (score > 0).astype(int)
def one_rep(rng):
    idx = rng.permutation(n); nt = n // 10
    te, tr = idx[:nt], idx[nt:]
    return np.mean(adaboost(X[tr], y[tr], X[te], rng) != y[te])
seeds = int(sys.argv[1]) if len(sys.argv) > 1 else 6
print("rows", n, "inputs", M, "test cases per repetition", n // 10, "trees", T)
res = []
for seed in range(seeds):
    rng = np.random.default_rng(seed)
    e = 100 * np.mean([one_rep(rng) for _ in range(100)])
    res.append(e); print(f"seed {seed}: mean test error over 100 repetitions = {e:.2f}%")
print(f"mean over {seeds} seeds {np.mean(res):.2f}%  sd across seeds {np.std(res, ddof=1):.2f}")

Expected output

rows 768 inputs 8 test cases per repetition 76 trees 50
seed 0: mean test error over 100 repetitions = 26.12%
seed 1: mean test error over 100 repetitions = 25.37%
seed 2: mean test error over 100 repetitions = 25.49%
seed 3: mean test error over 100 repetitions = 25.86%
seed 4: mean test error over 100 repetitions = 26.04%
seed 5: mean test error over 100 repetitions = 25.17%
mean over 6 seeds 25.67%  sd across seeds 0.39

Verifications (0)

Nobody has run it yet.