def buildTree(
dataset,
attrs,
treeNode=None,
):
if treeNode == None:
treeNode = TreeNode()
#determine if the cut attribute is 0
if len(attrs) <= 0:
#raise TypeError('len(attrs)<=0')
#if 0, find the max
label = findmaxlabel(dataset)
treeNode.setlabel(label)
return treeNode
#determine if pure value or not
tset = set()
sign = True
for row in dataset.getrows():
tset.add(row[-1])
if len(tset) > 1:
sign = False
break
if sign:
label = (list(tset))[0]
treeNode.setlabel(label)
return treeNode
#compute the gain
max_gain = 0.0
attr_id = None
entropys = calentropy1(dataset)
for attr in attrs:
attrentropy = calentropy2(attr, dataset)
gain = entropys - attrentropy
if gain > max_gain:
max_gain = gain
attr_id = attr
#vote to decide if no change in entropy after cut
if attr_id == None:
label = findmaxlabel(dataset)
treeNode.setlabel(label)
return treeNode
#set node value
treeNode.setsplistattrindex(attr_id)
treeNode.setsplitattr(dataset.gettitle()[attr_id])
#classify data
subdatasets = splitdataset(dataset, attr_id)
#release ram
del (dataset)
#delete the current cutting attribute
attrs.discard(attr_id)
#build branch, choose the biggest branch to build
keyset = set(subdatasets.keys())
for x in range(len(keyset)):
max_count = 0
rkey = None
for key in keyset:
value = subdatasets[key].getrowcount()
if value > max_count:
max_count = value
rkey = key
keyset.discard(rkey)
ctreenode = TreeNode(treeNode)
treeNode.children[rkey] = ctreenode
buildTree(subdatasets.get(rkey), attrs, ctreenode)
return treeNode
