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
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
| import numpy as np
class BaseDecisionTree:
def entropy(self, y):
_, cnt = np.unique(y, return_counts=True)
probs = cnt / y.shape[0]
return -np.sum(probs * np.log(probs) / np.log(2))
def gini(self, y):
_, cnt = np.unique(y, return_counts=True)
probs = cnt / y.shape[0]
return 1 - np.sum(np.square(probs))
def calcInfoGainRatio(self, y, y_sub):
n_samples = y.shape[0]
entropy = self.entropy(y)
new_entropy = np.sum([ny.size / n_samples * self.entropy(ny) for ny in y_sub])
gain = entropy - new_entropy
iv = - np.sum([ny.size / n_samples * np.log(ny.size / n_samples) for ny in y_sub])
gain_ratio = gain / iv
return gain_ratio
def calcInfoGain(self, y, y_sub):
n_samples = y.shape[0]
entropy = self.entropy(y)
new_entropy = np.sum([ny.size / n_samples * self.entropy(ny) for ny in y_sub])
return entropy - new_entropy
def calcGiniGain(self, y, y_sub):
n_samples = y.shape[0]
gini = self.gini(y)
new_gini = np.sum([ny.size / n_samples * self.gini(ny) for ny in y_sub])
return gini - new_gini
def storeTree(self):
import pickle
with open('DecisionTree.txt', 'wb') as fw:
pickle.dump(self.decisionTree, fw)
def loadTree(self):
import pickle
with open('DecisionTree.txt', 'rb') as fr:
self.decisionTree = pickle.load(fr)
class BinaryDecisionTree(BaseDecisionTree):
def __init__(self, loss='info_gain'):
self.loss = loss
self.decisionTree = {}
def fit(self, x, y):
self.decisionTree = self.createTree(x, y)
def createTree(self, x, y):
if np.unique(y).size == 1:
return y[0][0]
index, threshold = self.split(x, y)
if index == -1:
return np.argmax(np.bincount(y.flatten()))
decisionTree = {'split_index': index, 'threshold': threshold}
pos = x[:, index] < threshold
decisionTree['left'] = self.createTree(x[pos], y[pos])
decisionTree['right'] = self.createTree(x[~pos], y[~pos])
return decisionTree
def split(self, x, y):
n_samples, n_feat = x.shape[:]
opt_loss, split_index, threshold = 0, -1, None
for i in range(n_feat):
unique = np.unique(np.sort(x[:, i]))
split_val = (unique[:-1] + unique[1:]) / 2
for val in split_val:
pos = x[:, i] < val
y_sub = [y[pos], y[~pos]]
if self.loss == 'info_gain':
new_loss = self.calcInfoGain(y, y_sub)
elif self.loss == 'info_gain_ratio':
new_loss = self.calcInfoGainRatio(y, y_sub)
else:
new_loss = self.calcGiniGain(y, y_sub)
if new_loss > opt_loss:
opt_loss, split_index, threshold = new_loss, i, val
return split_index, threshold
def classify(self, inputs, node):
index = node['split_index']
nxt = node['left'] if inputs[index] < node['threshold'] else node['right']
if type(nxt).__name__ == 'dict':
return self.classify(inputs, nxt)
else:
return nxt
def predict(self, inputs):
res = [self.classify(sample, self.decisionTree) for sample in inputs]
return np.array(res)
class MultiwayDecisionTree(BaseDecisionTree):
"""
:param loss {"info_gain", "info_gain_ratio", "gini"}, default="info_gain"
"""
def __init__(self, loss='info_gain'):
self.loss = loss
self.decisionTree = {}
def fit(self, x, y):
used = np.full((x.shape[1], ), False)
self.decisionTree = self.createTree(x, y, used)
def createTree(self, x, y, used):
if np.unique(y).size == 1:
return y[0][0]
most_freq = np.argmax(np.bincount(y.flatten()))
if np.all(used):
return most_freq
index = self.split(x, y, used)
if index == -1:
return most_freq
used[index] = True
decisionTree = {'split_index': index, 'others': most_freq}
unique = np.unique(x[:, index])
for val in unique:
pos = np.where(x[:, index] == val)
decisionTree[val] = self.createTree(x[pos], y[pos], used)
return decisionTree
def split(self, x, y, used):
n_samples, n_feat = x.shape[:]
opt_loss, split_index = 0., -1
for i in range(n_feat):
if used[i]:
continue
unique = np.unique(x[:, i])
y_sub = [y[np.where(x[:, i] == val)] for val in unique]
if self.loss == 'info_gain':
new_loss = self.calcInfoGain(y, y_sub)
elif self.loss == 'info_gain_ratio':
new_loss = self.calcInfoGainRatio(y, y_sub)
else:
new_loss = self.calcGiniGain(y, y_sub)
if new_loss > opt_loss:
max_gain, split_index = new_loss, i
return split_index
def classify(self, inputs, node):
index = node['split_index']
if inputs[index] not in node:
return node['others']
nxt = node[inputs[index]]
if type(nxt).__name__ == 'dict':
return self.classify(inputs, nxt)
else:
return nxt
def predict(self, inputs):
res = [self.classify(sample, self.decisionTree) for sample in inputs]
return np.array(res)
|
