def buildStumpNode(self, now_root):
now_counts = self.getCount(now_root.data_index)
if now_counts == None or len(now_counts) == 0:
return
if len(now_root.data_index) <= self.max_leaf_data or len(now_counts) == 1:
now_root.is_leaf = True
self.setClassfy(now_root, now_counts)
return
best_value, best_sub_data_indexs = self.getSubIndex(now_root.data_index, self.tree_divide_feature_index)
for i in range(len(best_sub_data_indexs)):
new_son = Node(best_sub_data_indexs[i])
new_son.setDivideFeatureAndValue(self.tree_divide_feature_index, best_value[0] + best_value[1] * (i + 1))
self.buildStumpNode(new_son)
now_root.son.append(new_son)
def buildRandomNode(self, now_root):
now_counts = self.getCount(now_root.data_index)
if now_counts == None or len(now_counts) == 0:
return
if len(now_root.data_index) <= self.max_leaf_data or len(
now_counts) == 1:
now_root.is_leaf = True
self.setClassfy(now_root, now_counts)
return
best_sub_data_indexs = []
best_gain = -100000000
best_divide_feature_index = -1
best_value = ()
# 随机选择用于划分的特征值的子集
divide_list = list(range(self.feature_len))
random.shuffle(divide_list)
divide_list = divide_list[0:self.divide_feature_len]
for i in divide_list:
now_value, now_sub_data_indexs = self.getSubIndex(
now_root.data_index, i)
now_gain = self.getGain(now_root.data_index, now_sub_data_indexs)
if best_gain < now_gain:
best_divide_feature_index = i
best_gain = now_gain
best_sub_data_indexs = now_sub_data_indexs
best_value = now_value
for i in range(len(best_sub_data_indexs)):
new_son = Node(best_sub_data_indexs[i])
new_son.setDivideFeatureAndValue(
best_divide_feature_index,
best_value[0] + best_value[1] * (i + 1))
self.buildRandomNode(new_son)
now_root.son.append(new_son)