def learn_perceptron(training_set):
"""
@type training_set: list of DataElement
"""
omega = numpy.zeros(training_set[0].features.shape)
best_omega = omega.copy()
actual = [test.answer for test in training_set]
answers = [numpy.sign(omega.dot(test.features.T)) for test in training_set]
best_error_rate = error_rate(answers, actual)
for x in xrange(100):
updates = 0
for test in training_set:
if numpy.sign(omega.dot(test.features.T)) != test.answer:
updates += 1
omega += test.answer * test.features
answers, actual = get_answers_and_actual(training_set, omega)
new_error = error_rate(answers, actual)
if new_error < best_error_rate:
best_error_rate = new_error
best_omega = omega.copy()
if updates == 0:
break
return best_omega
def learn_perceptron(training_set):
"""
@type training_set: list of DataElement
"""
omega = numpy.zeros(training_set[0].features.shape)
best_omega = omega.copy()
actual = [test.answer for test in training_set]
answers = [numpy.sign(omega.dot(test.features.T)) for test in training_set]
best_error_rate = error_rate(answers, actual)
for x in xrange(100):
updates = 0
for test in training_set:
if numpy.sign(omega.dot(test.features.T)) != test.answer:
updates += 1
omega += test.answer * test.features
answers, actual = get_answers_and_actual(training_set, omega)
new_error = error_rate(answers, actual)
if new_error < best_error_rate:
best_error_rate = new_error
best_omega = omega.copy()
if updates == 0:
break
return best_omega
def main():
data_set = get_data_set()
error_rates = []
for x in xrange(40):
print "test %d" % (x + 1)
training_set, test_set = training_and_test_sets(data_set)
omega = learn_perceptron(training_set)
real_answers = [test.answer for test in test_set]
classification_answers = [numpy.sign(omega.dot(test.features.T)) for test in test_set]
rate = error_rate(classification_answers, real_answers)
error_rates.append(rate)
print "error_rate: %f" % rate
print "precision: %f" % precision(classification_answers, real_answers)
print "recall: %f" % recall(classification_answers, real_answers)
print "error mean error: %0.5f" % (sum(error_rates) / float(len(error_rates)))
def main():
data_set = get_data_set()
error_rates = []
for x in xrange(40):
print "test %d" % (x + 1)
training_set, test_set = training_and_test_sets(data_set)
omega = learn_perceptron(training_set)
real_answers = [test.answer for test in test_set]
classification_answers = [
numpy.sign(omega.dot(test.features.T)) for test in test_set
]
rate = error_rate(classification_answers, real_answers)
error_rates.append(rate)
print "error_rate: %f" % rate
print "precision: %f" % precision(classification_answers, real_answers)
print "recall: %f" % recall(classification_answers, real_answers)
print "error mean error: %0.5f" % (sum(error_rates) /
float(len(error_rates)))
