def ada_boosting(X, y, L):
D = [1.0/len(X)] * len(X)
ensemble = [] #{stump": => , "alpha" : => }
for i in range(L):
errorSum = 0
bestFeature, stump = decisionStump.train(X, y, D)
#calculate error sum
for i, (example, true_label) in enumerate(zip(X, y)):
if decisionStump.predict(example, stump, bestFeature) != true_label:
errorSum += D[i]
mean_error = errorSum / sum(D)
if mean_error == 0:
print >> sys.stderr, "achieved perfect training, exiting at iteration:", L
break
alpha = .5 * log((1 - mean_error) / mean_error)
#update weights
for i, (example, true_label) in enumerate(zip(X, y)):
if decisionStump.predict(example, stump, bestFeature) == true_label:
D[i] *= exp(-1 * alpha)
else:
D[i] *= exp(alpha)
ensemble.append({"stump": stump, "bestFeature": bestFeature, "alpha": alpha})
return ensemble
def create_bag(x, y):
sample_x, sample_y = zip(*sample(zip(x, y), num_samples))
sample_x, sample_y = np.array(sample_x), np.array(sample_y)
return decisionStump.train(sample_x, sample_y)