def classify_instance(boosted_tree, example):
trueVotes = 0.0
falseVotes = 0.0
for model in boosted_tree:
if dt.classify_sequence(model[0], example.sequence):
trueVotes += 1.0 * log(1 / model[1])
else:
falseVotes += 1.0 * log(1/ model[1])
return trueVotes >= falseVotes
def construct_boosted_tree(training):
weights = map(lambda x: 1.0 / len(training), training)
decisionTrees = []
for roundIdx in range(0, NUMBER_OF_ROUNDS):
weights_sum = sum(weights)
weights = map(lambda weight: weight / weights_sum, weights)
tree = dt.construct_tree(weighted_bootstrap_set(training, weights))
error = 0.0
for i in range(0, len(training)):
example = training[i]
if example.promoter != dt.classify_sequence(tree, example.sequence):
error += weights[i]
if error > .5:
return decisionTrees
beta = error / (1 - error)
for i in range(0, len(training)):
example = training[i]
if example.promoter == dt.classify_sequence(tree, example.sequence):
weights[i] = weights[i] * beta
decisionTrees.append((tree, beta))
return decisionTrees
def classify_instance(ensemble, example):
trueVotes = 0
for model in ensemble:
if dt.classify_sequence(model, example.sequence):
trueVotes += 1
return trueVotes > len(ensemble) / 2
