def maxCoverRule_wrong(cars_list, data_case):
for i in range(len(cars_list)):
if cars_list[i].class_label != data_case[-1]:
temp_data_case = data_case[:-1]
temp_data_case.append(cars_list[i].class_label)
if cba_cb_m1.is_satisfy(temp_data_case, cars_list[i]):
return i
return None
def allCoverRules(u, data_case, c_rule, cars_list):
w_set = set()
for rule_index in u:
# have higher precedences than cRule
if compare(cars_list[rule_index], c_rule) > 0:
# wrongly classify the data case
if cba_cb_m1.is_satisfy(data_case, cars_list[rule_index]) == False:
w_set.add(rule_index)
return w_set
def maxCoverRule_correct(cars_list, data_case):
for i in range(len(cars_list)):
if cars_list[i].class_label == data_case[-1]:
if cba_cb_m1.is_satisfy(data_case, cars_list[i]):
return i
return None
def classifier_builder_m2(cars, dataset):
classifier = Classifier_m2()
cars_list = cba_cb_m1.sort(cars)
for i in range(len(cars_list)):
cars_list[i] = ruleitem2rule(cars_list[i], dataset)
# stage 1
q = set()
u = set()
a = set()
mark_set = set()
for i in range(len(dataset)):
c_rule_index = maxCoverRule_correct(cars_list, dataset[i])
w_rule_index = maxCoverRule_wrong(cars_list, dataset[i])
if c_rule_index is not None:
u.add(c_rule_index)
if c_rule_index:
cars_list[c_rule_index].classCasesCovered[dataset[i][-1]] += 1
if c_rule_index and w_rule_index:
if compare(cars_list[c_rule_index], cars_list[w_rule_index]) > 0:
q.add(c_rule_index)
mark_set.add(c_rule_index)
else:
a.add((i, dataset[i][-1], c_rule_index, w_rule_index))
elif c_rule_index is None and w_rule_index is not None:
a.add((i, dataset[i][-1], c_rule_index, w_rule_index))
# stage 2
for entry in a:
if cars_list[entry[3]] in mark_set:
if entry[2] is not None:
cars_list[entry[2]].classCasesCovered[entry[1]] -= 1
cars_list[entry[3]].classCasesCovered[entry[1]] += 1
else:
if entry[2] is not None:
w_set = allCoverRules(u, dataset[entry[0]],
cars_list[entry[2]], cars_list)
else:
w_set = allCoverRules(u, dataset[entry[0]], None, cars_list)
for w in w_set:
cars_list[w].replace.add((entry[2], entry[0], entry[1]))
cars_list[w].classCasesCovered[entry[1]] += 1
q |= w_set
# stage 3
rule_errors = 0
q = sort_with_index(q, cars_list)
data_cases_covered = list([False] * len(dataset))
for r_index in q:
if cars_list[r_index].classCasesCovered[
cars_list[r_index].class_label] != 0:
for entry in cars_list[r_index].replace:
if data_cases_covered[entry[1]]:
cars_list[r_index].classCasesCovered[entry[2]] -= 1
else:
if entry[0] is not None:
cars_list[entry[0]].classCasesCovered[entry[2]] -= 1
for i in range(len(dataset)):
datacase = dataset[i]
if datacase:
is_satisfy_value = cba_cb_m1.is_satisfy(
datacase, cars_list[r_index])
if is_satisfy_value:
dataset[i] = []
data_cases_covered[i] = True
rule_errors += errorsOfRule(cars_list[r_index], dataset)
class_distribution = compClassDistr(dataset)
default_class = selectDefault(class_distribution)
default_errors = defErr(default_class, class_distribution)
total_errors = rule_errors + default_errors
classifier.add(cars_list[r_index], default_class, total_errors)
classifier.discard()
return classifier