def normalizer(D, A):
sum = 0.0
for v in D.get_attributeValues(A):
D_j = Dataset(None, None, None, D, A, v)
pr_A = D_j.get_dataSize() / D.get_dataSize()
sum += pr_A * math.log(pr_A, 2)
sum = -1 * sum
#print('Normalizer: %.3f' % (sum))
return sum
def entropy(D, A=None):
sum = 0.0
if A:
for v_j in D.get_attributeValues(A):
D_j = Dataset(None, None, None, D, A, v_j)
sum += D_j.get_dataSize() / D.get_dataSize() * entropy(D_j)
#print('Entropy[%s]: %.3f' % (A, sum))
else:
for c_j in D.get_classes():
if D.pr_c(c_j):
sum += D.pr_c(c_j) * math.log(D.pr_c(c_j), 2)
sum = -1 * sum
#print('Entropy: %.3f' % (sum))
return sum
def decision_tree_rec(self, D, A, T, threshold):
assert D.get_numClasses() > 0
if D.get_numClasses() == 1:
print('make T a leaf node with labeled with c');
classification = D.get_classes().pop()
print('F**K')
print(classification)
num, choice = D.get_num_choice_tuple(D.get_classAttribute(), classification)
decision = ElementTree.SubElement(T, 'decision')
decision.set('end', '1')
decision.set('num', num)
decision.set('choice', choice)
elif len(A) == 0:
print('make T a leaf node labeled with the most frequent class')
classification = D.get_mostPluralClass()
num, choice = D.get_num_choice_tuple(D.get_classAttribute(), classification)
decision = ElementTree.SubElement(T, 'decision')
decision.set('end', '1')
decision.set('num', num)
decision.set('choice', choice)
else:
print('contains examples belonging to a mixture of classes')
A_split = self.select_splitting_attribute(D, A, threshold)
print('SPLITTING ON %s: ', A_split)
#A_split = select_splitting_attribute_ratio(D, A, threshold)
#print('SPLITTING ON RATIO %s: ', A_split)
if A_split == None:
decision = ElementTree.SubElement(T, 'decision')
decision.set('end', '1')
decision.set('choice', D.get_mostPluralClass())
else:
node = ElementTree.SubElement(T, 'node')
node.set('var', A_split)
AminusA_split = set()
for a in A:
if a != A:
AminusA_split.add(a)
for v in D.get_attributeValues(A_split):
D_v = Dataset(None, None, None, D, A_split, v)
if D_v.get_dataSize() > 0:
edge = ElementTree.SubElement(node, 'edge')
num, var = D_v.get_num_choice_tuple(A_split, v)
edge.set('var', var)
edge.set('num', num)
self.decision_tree_rec(D_v, AminusA_split, edge, threshold)
else:
print('IGNORING %s' % str(v))
def __init__(self,
domain_filename, csv_filename, restrictions_filename = None,
ratio=False
):
print('~' * 20)
print(csv_filename)
print(restrictions_filename)
print(ratio)
self.tree = None
self.tree_name = csv_filename[:-4]
self.tree_name = os.path.splitext(csv_filename)[0]
self.restricted = False
if restrictions_filename:
print('F*****G RESTRICTED')
self.restricted = True
else:
print('NOT RESTRICTED?!?! %s' % restrictions_filename)
print('~' * 20)
self.select_splitting_attribute = select_splitting_attribute_default
if ratio:
self.select_splitting_attribute = select_splitting_attribute_ratio
self.trainingSet = Dataset(domain_filename, csv_filename, restrictions_filename)
class DecisionTreeBuilder(object):
def __init__(self,
domain_filename, csv_filename, restrictions_filename = None,
ratio=False
):
print('~' * 20)
print(csv_filename)
print(restrictions_filename)
print(ratio)
self.tree = None
self.tree_name = csv_filename[:-4]
self.tree_name = os.path.splitext(csv_filename)[0]
self.restricted = False
if restrictions_filename:
print('F*****G RESTRICTED')
self.restricted = True
else:
print('NOT RESTRICTED?!?! %s' % restrictions_filename)
print('~' * 20)
self.select_splitting_attribute = select_splitting_attribute_default
if ratio:
self.select_splitting_attribute = select_splitting_attribute_ratio
self.trainingSet = Dataset(domain_filename, csv_filename, restrictions_filename)
def decision_tree_rec(self, D, A, T, threshold):
assert D.get_numClasses() > 0
if D.get_numClasses() == 1:
print('make T a leaf node with labeled with c');
classification = D.get_classes().pop()
print('F**K')
print(classification)
num, choice = D.get_num_choice_tuple(D.get_classAttribute(), classification)
decision = ElementTree.SubElement(T, 'decision')
decision.set('end', '1')
decision.set('num', num)
decision.set('choice', choice)
elif len(A) == 0:
print('make T a leaf node labeled with the most frequent class')
classification = D.get_mostPluralClass()
num, choice = D.get_num_choice_tuple(D.get_classAttribute(), classification)
decision = ElementTree.SubElement(T, 'decision')
decision.set('end', '1')
decision.set('num', num)
decision.set('choice', choice)
else:
print('contains examples belonging to a mixture of classes')
A_split = self.select_splitting_attribute(D, A, threshold)
print('SPLITTING ON %s: ', A_split)
#A_split = select_splitting_attribute_ratio(D, A, threshold)
#print('SPLITTING ON RATIO %s: ', A_split)
if A_split == None:
decision = ElementTree.SubElement(T, 'decision')
decision.set('end', '1')
decision.set('choice', D.get_mostPluralClass())
else:
node = ElementTree.SubElement(T, 'node')
node.set('var', A_split)
AminusA_split = set()
for a in A:
if a != A:
AminusA_split.add(a)
for v in D.get_attributeValues(A_split):
D_v = Dataset(None, None, None, D, A_split, v)
if D_v.get_dataSize() > 0:
edge = ElementTree.SubElement(node, 'edge')
num, var = D_v.get_num_choice_tuple(A_split, v)
edge.set('var', var)
edge.set('num', num)
self.decision_tree_rec(D_v, AminusA_split, edge, threshold)
else:
print('IGNORING %s' % str(v))
def build_tree(self, threshold):
self.tree = Element('Tree')
self.tree.set('name',self.tree_name)
allAttributes = self.trainingSet.get_attributes()
self.decision_tree_rec(self.trainingSet, allAttributes, self.tree, threshold)
return self.tree
def get_tree(self):
return self.tree
def get_xml(self, indent=' '):
return minidom.parseString(
ElementTree.tostring(self.tree)).toprettyxml(indent=indent)
def print_tree(self, file=sys.stdout, indent=' '):
xml_str = self.get_xml(indent)
print(xml_str, file=file)
return xml_str
def save_tree(self, file=None, indent=' '):
if file:
return self.print_tree(file, indent)
else:
xml_filename = self.tree_name
if self.restricted:
xml_filename += '_restricted'
xml_filename += '.xml'
with open(self.tree_name + '.xml', 'w') as save_file:
xml_str = self.print_tree(save_file, indent)
return xml_str
