def rule_generator(dataset, minsup, minconf):
frequent_ruleitems = FrequentRuleitems()
car = Car()
# get large 1-ruleitems and generate rules
class_label = set([x[-1] for x in dataset])
for column in range(0, len(dataset[0]) - 1):
distinct_value = set([x[column] for x in dataset])
for value in distinct_value:
cond_set = {column: value}
for classes in class_label:
rule_item = ruleitem.RuleItem(cond_set, classes, dataset)
if rule_item.support >= minsup:
frequent_ruleitems.add(rule_item)
car.gen_rules(frequent_ruleitems, minsup, minconf)
cars = car
last_cars_number = 0
current_cars_number = len(cars.rules)
while frequent_ruleitems.get_size() > 0 and current_cars_number <= 2000 and \
(current_cars_number - last_cars_number) >= 10:
candidate = candidate_gen(frequent_ruleitems, dataset)
frequent_ruleitems = FrequentRuleitems()
car = Car()
for item in candidate.frequent_ruleitems_set:
if item.support >= minsup:
frequent_ruleitems.add(item)
car.gen_rules(frequent_ruleitems, minsup, minconf)
cars.append(car, minsup, minconf)
last_cars_number = current_cars_number
current_cars_number = len(cars.rules)
return cars
def join(item1, item2, dataset):
if item1.class_label != item2.class_label:
return None
category1 = set(item1.cond_set)
category2 = set(item2.cond_set)
if category1 == category2:
return None
intersect = category1 & category2
for item in intersect:
if item1.cond_set[item] != item2.cond_set[item]:
return None
category = category1 | category2
new_cond_set = dict()
for item in category:
if item in category1:
new_cond_set[item] = item1.cond_set[item]
else:
new_cond_set[item] = item2.cond_set[item]
new_ruleitem = ruleitem.RuleItem(new_cond_set, item1.class_label, dataset)
return new_ruleitem
def prune(rule, dataset):
import sys
min_rule_error = sys.maxsize
pruned_rule = rule
# prune rule recursively
def find_prune_rule(this_rule):
nonlocal min_rule_error
nonlocal pruned_rule
# calculate how many errors the rule r make in the dataset
def errors_of_rule(r):
import apr_cb_m1
errors_number = 0
for case in dataset:
if apr_cb_m1.is_satisfy(case, r) == False:
errors_number += 1
return errors_number
rule_error = errors_of_rule(this_rule)
if rule_error < min_rule_error:
min_rule_error = rule_error
pruned_rule = this_rule
this_rule_cond_set = list(this_rule.cond_set)
if len(this_rule_cond_set) >= 2:
for attribute in this_rule_cond_set:
temp_cond_set = dict(this_rule.cond_set)
temp_cond_set.pop(attribute)
temp_rule = ruleitem.RuleItem(temp_cond_set,
this_rule.class_label, dataset)
temp_rule_error = errors_of_rule(temp_rule)
if temp_rule_error <= min_rule_error:
min_rule_error = temp_rule_error
pruned_rule = temp_rule
if len(temp_cond_set) >= 2:
find_prune_rule(temp_rule)