def __init__(self, lst_sample, lst_feature):
self.lst_feature = lst_feature
self.lst_sample = lst_sample
self.n_sample = len(self.lst_sample)
self.left_child = None
self.right_child = None
self.split_fea = None
self.split_fea_val = 0
self.split_fea_index = 0
self.split_fea_index_tmp = 0
self.is_leaf = True
self.predict_val = mean([sample[-1] for sample in self.lst_sample])
'''
def try_split_node(self, criteria = 'MSE'):
if len(self.lst_sample) <= 1:
raise ValueError("Empty lst_sample! Can NOT split any more!")
if criteria == "MSE":
mse_min = sys.maxint
self.split_fea_index = -1
self.split_fea_val = 0
self.left_child = None
self.right_child = None
for index_fea in range(len(self.lst_sample[0]) - 1):
lst_flt_split = [sample[index_fea] for sample in self.lst_sample]
max_flt_split = max(lst_flt_split)
min_flt_split = min(lst_flt_split)
set_flt_split = set()
for sample_split in self.lst_sample:
lst_sample_left = []
lst_sample_right = []
flt_split = sample_split[index_fea]
# If the split feature value is the max or min or has been evaluated before, then skip
if flt_split == max_flt_split or flt_split == min_flt_split or flt_split in set_flt_split:
continue
set_flt_split.add(flt_split)
for sample in self.lst_sample:
if sample[index_fea] <= flt_split:
lst_sample_left.append(sample)
else:
lst_sample_right.append(sample)
mean_left = mean([sample[-1] for sample in lst_sample_left])
mean_right = mean([sample[-1] for sample in lst_sample_right])
mse = sum([(sample[-1] - mean_left) ** 2 for sample in lst_sample_left]) + sum([(sample[-1] - mean_right) ** 2 for sample in lst_sample_right])
if mse < mse_min:
mse_min = mse
self.split_fea_index_tmp = index_fea
self.split_fea_val_tmp = sample_split[index_fea]
self.left_child_tmp = tree_node(lst_sample_left, self.lst_feature)
self.right_child_tmp = tree_node(lst_sample_right, self.lst_feature)
return (mse_min, self.split_fea_index_tmp, self.split_fea_val_tmp, self.left_child_tmp, self.right_child_tmp)
elif criteria == "IG":
pass
