def runAdult():
from data import adult
from decisionstump import buildDecisionStump
train, test = adult.load()
weakLearner = buildDecisionStump
rounds = 20
h = boosting.boost(train, weakLearner, rounds)
print("Training error: %G" % error(h, train))
print("Test error: %G" % error(h, test))
def runAdult():
from data import adult
from decisionstump import buildDecisionStump
train, test = adult.load()
weakLearner = buildDecisionStump
rounds = 20
h = boosting.boost(train, weakLearner, rounds)
print("Training error: %G" % error(h, train))
print("Test error: %G" % error(h, test))
def statistics(train, test, protectedIndex, protectedValue, numRounds=20): weight = 0.5 flipProportion = 0.2 error = makeErrorFunction(protectedIndex, protectedValue, weight) weakLearner = lambda draw: buildDecisionStump(draw, errorFunction=error) h = boosting.boost(train, weakLearner=weakLearner) bias = ef.signedStatisticalParity(test, protectedIndex, protectedValue, h) error = ef.labelError(test, h) ubif = ef.individualFairness(train, boosting.boost, flipProportion) return error, bias, ubif
def statistics(train, test, protectedIndex, protectedValue, numRounds=20): weight = 0.5 flipProportion = 0.2 error = makeErrorFunction(protectedIndex, protectedValue, weight) weakLearner = lambda draw: buildDecisionStump(draw, errorFunction=error) h = boosting.boost(train, weakLearner = weakLearner) bias = ef.signedStatisticalParity(test, protectedIndex, protectedValue, h) error = ef.labelError(test, h) ubif = ef.individualFairness(train, boosting.boost, flipProportion) return error, bias, ubif
def simpleTest():
def target(x):
if x[2] > 0.5 or x[3] > 0.5:
return 1 if random.random() > 0.05 else -1
return -1
examples = [[random.random() for _ in range(10)] for _ in range(1000)]
labels = [target(x) for x in examples]
trainingData = list(zip(examples, labels))
testData = [[random.random() for _ in range(10)] for _ in range(1000)]
testLabels = [target(x) for x in testData]
def testCoordinate(samples, j):
values = [sign(x[j] - 0.5) * y for (x,y) in samples]
return len([z for z in values if z > 0]) / len(values)
def bestCoordinate(samples, n):
return max(range(n), key=lambda j: testCoordinate(samples, j))
# find the single coordinate and a threshold value that works best
def singleCoordinateLearner(drawExample):
samples = [drawExample() for _ in range(100)]
n = len(samples[0][0])
j = bestCoordinate(samples, n)
return lambda x: x[j] > 0.5
finalH, finalDistr = boosting.boost(trainingData, singleCoordinateLearner, 100)
finalError = len([x for x in testData if finalH(x) != target(x)]) / len(testData)
print(finalError)
def boostingLearner(data, protectedIndex, protectedValue):
h = boosting.boost(data)
return randomOneSideRelabelData(h, data, protectedIndex, protectedValue)
nonfavored_data = [(feats, label) for feats, label in trainingData
if not feats[protectedIndex] == favored_trait]
NF, NFn = (len(nonfavored_data),
len([1 for x, label in nonfavored_data if h(x) == -1]))
p = NF * abs(bias) / NFn
def relabeledClassifier(point):
origClass = h(point)
if point[protectedIndex] != favored_trait and origClass == -1:
if random() < p:
return -origClass
else:
return origClass
else:
return origClass
return relabeledClassifier
if __name__ == '__main__':
from data import adult
from boosting import boost
trainingData, testData = adult.load()
protectedIndex = adult.protectedIndex
protectedValue = adult.protectedValue
h = boost(trainingData, 5)
rr = randomOneSideRelabelData(h, trainingData, protectedIndex,
protectedValue)
# to get rated 1
def randomOneSideRelabelData(h, trainingData, protectedIndex, protectedValue):
bias = signedStatisticalParity(trainingData, protectedIndex, protectedValue, h)
favored_trait = zeroOneSign(bias)
nonfavored_data = [(feats,label) for feats,label in trainingData if not feats[protectedIndex]==favored_trait]
NF, NFn = len(nonfavored_data), len([1 for x,label in nonfavored_data if h(x)==-1])
p = NF*abs(bias)/NFn
def relabeledClassifier(point):
origClass = h(point)
if point[protectedIndex] != favored_trait and origClass == -1:
if random() < p:
return -origClass
else:
return origClass
else:
return origClass
return relabeledClassifier
if __name__ == '__main__':
from data import adult
from boosting import boost
trainingData, testData = adult.load()
protectedIndex = adult.protectedIndex
protectedValue = adult.protectedValue
h = boost(trainingData, 5)
rr = randomOneSideRelabelData(h, trainingData, protectedIndex, protectedValue)
def boostingLearner(data, protectedIndex, protectedValue): h = boosting.boost(data) return randomOneSideRelabelData(h, data, protectedIndex, protectedValue)
def boostingLearner(data, protectedIndex, protectedValue):
return boost(data)
if label != row[-1]:
numErrors = numErrors + 1
print(float(numErrors) / len(perTestMatrix))
# Boosting Example Use
# get training data
boostTrainingData = open('data/boosttrain.txt', 'r').readlines()
boostTrainingMatrix = []
for line in boostTrainingData:
boostTrainingMatrix.append(np.fromstring(line, dtype=float, sep=' '))
boostTrainingMatrix = np.array(boostTrainingMatrix)
# get test data
boostTestData = open('data/boosttest.txt', 'r').readlines()
boostTestMatrix = []
for line in boostTestData:
boostTestMatrix.append(np.fromstring(line, dtype=float, sep=' '))
boostTestMatrix = np.array(boostTestMatrix)
# get an array of weak learners
f = boost(boostTrainingMatrix, 4)
# get test error
numErrors = 0
for row in boostTestMatrix:
label = strongLearner(f, row)
if label != row[-1]:
numErrors = numErrors + 1
print(float(numErrors) / len(boostTestMatrix))
def boostingLearner(data, protectedIndex, protectedValue):
return boost(data)