-
Notifications
You must be signed in to change notification settings - Fork 0
/
decision_tree.py
251 lines (224 loc) · 9.04 KB
/
decision_tree.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
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
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
import sys
import math
import pickle
import copy
import pandas as pd
class DecisionTree:
def __init__(self, training_data=None, col_with_category=None, path_to_file=None):
self.leaf_counter = 0
self.current_node_id = 0
if path_to_file:
infile = open(path_to_file, 'rb')
self.__dict__.update(pickle.load(infile).__dict__)
infile.close()
else:
self.training_data = training_data
self.col_with_category = col_with_category
print('Building tree ...')
self.root = self.__build_tree(training_data)
def __str__(self):
map = self.__traverse(self.root)
leafes = 0
for row in map:
for element in row:
if type(element).__name__ == 'Leaf':
leafes += 1
return f'Decision tree built from {len(self.training_data)} elements, leafs: {leafes}'
def trim_tree(self, data):
tree = self
processed_levels = 0
while True:
print('! Trimmed !')
temp, changed, processed_levels = tree.__trim_one_node(
data, processed_levels=processed_levels)
if not changed:
break
tree = temp
return tree
def test_accuracy(self, test_data):
pos = 0
for index, row in test_data.iterrows():
if self.find_category(row) == row[self.col_with_category]:
pos += 1
return pos/len(test_data)
def find_category(self, row):
current_node = self.root
while(type(current_node).__name__ != 'Leaf'):
current_node = current_node.get_child_node(
more_or_equal=row[current_node.attribute] >= current_node.value)
return current_node.category
def save_to_file(self, path):
outfile = open(path, 'wb')
pickle.dump(self, outfile)
outfile.close()
def __build_tree(self, data):
print(f'build tree from {len(data)} elements')
self.current_node_id += 1
if self.__stop_criteria(data):
tree = Leaf(self.current_node_id, category=self.__get_category(
data), n_of_elements=len(data))
self.leaf_counter += 1
else:
tests = self.__generate_test_pool(data)
results = ([], [])
while len(results[0]) == 0 or len(results[1]) == 0:
choosen_test = self.__choose_test(tests)
results = self.__split_by_attribute(data, choosen_test)
tree = Node(self.current_node_id, attribute=choosen_test[0], value=choosen_test[1])
tree.set_child_node(more_or_equal=True, node=self.__build_tree(results[0]))
tree.set_child_node(more_or_equal=False, node=self.__build_tree(results[1]))
return tree
def __test_quality(self, data, attr, split_point):
subsets = self.__split_by_attribute(data, (attr, split_point, 0))
sum0 = len(subsets[1])
sum1 = len(subsets[0])
suk0 = subsets[1][self.col_with_category].sum()
suk1 = subsets[0][self.col_with_category].sum()
if (sum0 == 0):
E_0 = 0
elif (suk0 == 0 or suk0 == sum0):
E_0 = 0
else:
E_0 = -(suk0/sum0)*math.log10(suk0/sum0) - \
((sum0-suk0)/sum0)*math.log10((sum0-suk0)/sum0)
if (sum1 == 0):
E_1 = 0
elif (suk1 == 0 or suk1 == sum1):
E_1 = 0
else:
E_1 = -(suk1/sum1)*math.log10(suk1/sum1) - \
((sum1-suk1)/sum1)*math.log10((sum1-suk1)/sum1)
E_w = sum0/len(data) * E_0 + sum1/len(data) * E_1
return(E_w)
def __stop_criteria(self, data):
if len(data) == 0 or data[self.col_with_category].nunique() == 1:
return True
return False
def __split_by_attribute(self, data, test):
subset1 = data[data[test[0]] >= test[1]]
subset2 = data[data[test[0]] < test[1]]
return (subset1, subset2)
def __get_category(self, data):
if len(data) == 0:
return -1
sum = data[self.col_with_category].sum()
if sum/len(data) >= 0.5:
return 1
else:
return 0
def __generate_test(self, data, attr, min, max):
mid = (min + max) / 2
entropy1 = self.__test_quality(data, attr, mid)
entropy2 = self.__test_quality(data, attr, (min+mid)/2)
entropy3 = self.__test_quality(data, attr, (max+mid)/2)
if entropy1 <= entropy2 and entropy1 <= entropy3:
split_point = mid
result_entropy = entropy1
elif entropy2 <= entropy3:
split_point, result_entropy = self.__generate_test(
data, attr, min, mid)
else:
split_point, result_entropy = self.__generate_test(
data, attr, mid, max)
return split_point, result_entropy
def __generate_test_pool(self, data):
result = []
for c in data.columns:
if c != self.col_with_category:
split_point, result_entropy = self.__generate_test(
data, c, data[c].min(), data[c].max())
result.append((c, split_point, result_entropy))
return result
def __choose_test(self, tests):
tests_df = pd.DataFrame(
tests, columns=['index', 'split_point', 'entropy'])
weights = tests_df.loc[:, 'entropy'].values.tolist()
for i in range(len(weights)):
if weights[i] == 0:
weights[i] = sys.maxsize * 2 + 1
else:
weights[i] = 1 / weights[i]
chosen_test_df = tests_df.sample(n=1, weights=weights)
chosen_test = chosen_test_df.iloc[0].to_numpy()
return chosen_test
def __traverse(self, rootnode, id=None):
tree_map = []
thislevel = [rootnode]
while thislevel:
nextlevel = list()
for n in thislevel:
if n.node_id == id:
return n
if type(n).__name__ != 'Leaf':
nextlevel.append(n.get_child_node(True))
nextlevel.append(n.get_child_node(False))
tree_map.append(thislevel)
thislevel = nextlevel
return tree_map
def __tree_to_leaf(self, rootnode):
categories = {}
thislevel = [rootnode]
while thislevel:
nextlevel = list()
for n in thislevel:
if type(n).__name__ != 'Leaf':
nextlevel.append(n.get_child_node(True))
nextlevel.append(n.get_child_node(False))
else:
if n.category not in categories:
categories[n.category] = 0
categories[n.category] += n.n_of_elements
thislevel = nextlevel
winning_cat = max(categories, key=categories.get)
leaf = Leaf(-1, category=winning_cat,
n_of_elements=sum(categories.values()))
return leaf
def __trim_one_node(self, data, processed_levels):
trimed_tree = copy.deepcopy(self)
trimed_tree_map = self.__traverse(trimed_tree.root)
row_number = 0
for row in trimed_tree_map:
row_number += 1
if row_number < processed_levels:
continue
for node in row:
if type(node).__name__ != 'Leaf':
print(f'Working on {type(node).__name__} {node.node_id}')
for side in [True, False]:
temp_node = node.get_child_node(side)
if type(temp_node).__name__ == 'Leaf':
continue
node.set_child_node(side, self.__tree_to_leaf(
node.get_child_node(side)))
old_acc = self.test_accuracy(data)
new_acc = trimed_tree.test_accuracy(data)
print(
f'\tFor {side} branch:\t old={old_acc} new={new_acc} better? {new_acc>=old_acc}')
if new_acc >= old_acc:
print(f'Trimming {side} side')
return trimed_tree, True, processed_levels
node.set_child_node(side, temp_node)
processed_levels += 1
return None, False, 0
class Node():
def __init__(self, node_id, attribute=None, value=None):
self.node_id = node_id
self.attribute = attribute
self.value = value
self.__more_eq_node = None
self.__less_node = None
def set_child_node(self, more_or_equal, node):
if more_or_equal:
self.__more_eq_node = node
else:
self.__less_node = node
def get_child_node(self, more_or_equal):
if more_or_equal:
return self.__more_eq_node
else:
return self.__less_node
class Leaf():
def __init__(self, node_id, category=None, n_of_elements=0):
self.node_id = node_id
self.category = category
self.n_of_elements = n_of_elements