-
Notifications
You must be signed in to change notification settings - Fork 0
/
ccp_prune.py
157 lines (125 loc) · 4.36 KB
/
ccp_prune.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
from copy import deepcopy
def gini(li):
n = len(li)
m = sum([i[-1] for i in li])
return m * (n - m) / n / n
def count_majority(li):
return 2 * sum([i[-1] for i in li]) > len(li)
class Node(object):
def __init__(self, id=None, name=None, key=None, left=None, right=None, label=None, leaf_sum=0, Rt=None, RT=None, majority=None):
self.feature_id = id
self.feature_name = name
self.split_value = key
self.left = left
self.right = right
self.leaf_sum = leaf_sum
self.label = label
self.Rt = Rt
self.RT = RT
self.majority = majority
class Tree(object):
def __init__(self, features):
self.features = features
self.root = None
def build(self, X, dep, max_dep, N):
def unique(li):
for i in li:
if i[-1] != li[0][-1]:
return False
return True
if unique(X) or dep == max_dep:
return Node(label=X[0][-1], leaf_sum=1, Rt=0, RT=0)
min_feature = None
min_key = None
min_Imp = 0
min_id = None
left = None
right = None
for i, name in enumerate(self.features):
X.sort(key=lambda x: x[i])
n = len(X)
for j in range(n - 1):
delta = gini(X) - (j + 1)/n * \
gini(X[:j+1]) - (n - j - 1)/n * gini(X[j+1:])
if delta > min_Imp and X[j][i] != X[j+1][i]:
min_Imp = delta
min_feature = name
min_id = i
min_key = (X[j][i] + X[j+1][i]) / 2
left = X[:j+1]
right = X[j+1:]
left = self.build(left, dep+1, max_dep, N)
right = self.build(right, dep+1, max_dep, N)
return Node(min_id, min_feature, min_key, left, right, None, left.leaf_sum+right.leaf_sum, gini(X)*len(X)*len(X)/N, left.RT+right.RT, count_majority(X))
def prune(t, alpha):
if t.label != None:
return
if t.RT + alpha * t.leaf_sum >= t.Rt + alpha:
t.label = t.majority
return
prune(t.left, alpha)
prune(t.right, alpha)
def tree_size(t):
if t.label != None:
return 1
return tree_size(t.left) + tree_size(t.right)
def cost_complexity_algorithm(tree, X):
alphas = {0}
def travel(t):
if t.label != None:
return
alphas.add((t.Rt - t.RT)/(t.leaf_sum-1))
travel(t.left)
travel(t.right)
travel(tree.root)
alphas = sorted(alphas)
pruned_trees = []
for a in alphas:
new_tree = deepcopy(tree)
prune(new_tree.root, a)
pruned_trees.append(new_tree)
return pruned_trees
def shuffle(li, seed):
for i in range(50):
li.remove(li[seed])
li.append(li[seed])
def process(features, X, seed):
confusion_matrix = [[0, 0], [0, 0]]
def classification(t, test_data):
if t.label != None:
res = 0
for i in test_data:
res += i[-1] == t.label
confusion_matrix[int(t.label)][int(i[-1])] += 1
return res
left = []
right = []
for i in test_data:
if i[t.feature_id] < t.split_value:
left.append(i)
else:
right.append(i)
return classification(t.left, left) + classification(t.right, right)
best_tree = None
best_score = 100000000
for k in range(0,100,10):
shuffle(X, seed + k)
mid = int(len(X) * 0.5)
my_tree = Tree(features)
my_tree.root = my_tree.build(
X[:mid], 0, 10, len(X[:mid]))
pruned_trees = cost_complexity_algorithm(my_tree, X[:mid])
# best_tree = pruned_trees[0]
# best_score = classification(best_tree.root, X[mid:]) / len(X[mid:])
for t in pruned_trees[1:]:
confusion_matrix = [[0, 0], [0, 0]]
score = classification(t.root, X[mid:]) / len(X[mid:])
# cishu = classification(t.root, X[mid:])
# if confusion_matrix[1][1] == 0:
# confusion_matrix[1][1] = 0.0001
# score = (confusion_matrix[1][0] + confusion_matrix[0]
# [1])/(confusion_matrix[1][1]**2)/(cishu**0.3)
if score < best_score:
best_tree = t
best_score = score
return best_tree