-
Notifications
You must be signed in to change notification settings - Fork 0
/
d_tree.py
153 lines (127 loc) · 5.6 KB
/
d_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
import pandas as pd
import numpy as np
from typing import Union, Optional
class Leaf:
def __init__(self, value: float) -> None:
self.value = value
def __repr__(self):
class_name = type(self).__name__
return f"{class_name}({self.value})"
def __str__(self):
return self.__repr__()
class TreeNode:
def __init__(self, col_name: str, split_value: float, ig: float,
left: Union["TreeNode", Leaf, None] = None, right: Union["TreeNode", Leaf, None] = None) -> None:
self.col_name = col_name
self.split_value = split_value
self.ig = ig
self.left = left
self.right = right
def __repr__(self):
return (f"TreeNode({self.col_name}, {self.split_value}, {self.ig}, "
f"{self.left.__repr__()}, {self.right.__repr__()})")
def __str__(self):
result = ""
stack = [(self, 0, "root")]
while stack:
node, indent, side = stack.pop()
if isinstance(node, TreeNode):
result += f"{' ' * indent}{node.col_name} > {node.split_value:.8f} | IG: {node.ig:.8f}\n"
stack.append((node.right, indent + 2, 'right'))
stack.append((node.left, indent + 2, 'left'))
elif isinstance(node, Leaf):
result += f"{' ' * indent}leaf_{side} = {node.value}\n"
else:
result += f"{' ' * indent}<UNKNOWN NODE>\n"
return result
class MyTreeClf:
def __init__(self, max_depth: int = 5, min_samples_split: int = 2, max_leafs: int = 20):
self.max_depth = max_depth
self.min_samples_split = min_samples_split
self.max_leaves = max_leafs if max_leafs >= 2 else 2
self.leaves_cnt = 0
self._splits_cnt = 1 # imply that root should be in any case
self.tree: Optional[TreeNode] = None
def __str__(self) -> str:
return f"MyTreeClf class: max_depth={self.max_depth}, min_samples_split={self.min_samples_split}, "\
f"max_leafs={self.max_leaves}"
@staticmethod
def _calculate_entropy(y: pd.Series) -> float:
p = y.value_counts() / len(y)
return -(p * p.apply(np.log2)).sum()
def _get_best_split(self, X, y) -> Union[TreeNode, Leaf]:
s0 = self._calculate_entropy(y)
N = len(y)
if len(y.value_counts()) == 1:
self.leaves_cnt += 1
return Leaf(float(y.iloc[0]))
col_name, split_value, ig = '', float('-inf'), 0
for col in X.columns:
tmp = pd.Series(X[col].sort_values().unique()).rolling(2).mean()
for split in tmp.iloc[1:]:
flt = X[col] <= split
part1, part2 = y.loc[flt], y.loc[~flt]
s1, s2 = self._calculate_entropy(part1), self._calculate_entropy(part2)
new_ig = s0 - s1 * len(part1) / N - s2 * len(part2) / N
if new_ig > ig:
col_name, split_value, ig = col, split, new_ig
return TreeNode(col_name, split_value, ig)
def _build_tree(self, X, y, node, depth) -> None:
flt = X[node.col_name] <= node.split_value
X_l, y_l = X[flt], y[flt]
X_r, y_r = X[~flt], y[~flt]
if depth >= self.max_depth or self._splits_cnt + 1 == self.max_leaves:
self.leaves_cnt += 2
node.left = Leaf(y_l.mean())
node.right = Leaf(y_r.mean())
else:
if len(y_l) < self.min_samples_split or self._splits_cnt + 1 == self.max_leaves:
self.leaves_cnt += 1
node.left = Leaf(y_l.mean())
else:
node.left = self._get_best_split(X_l, y_l)
if isinstance(node.left, TreeNode):
self._splits_cnt += 1
self._build_tree(X_l, y_l, node.left, depth + 1)
if len(y_r) < self.min_samples_split or self._splits_cnt + 1 == self.max_leaves:
self.leaves_cnt += 1
node.right = Leaf(y_r.mean())
else:
node.right = self._get_best_split(X_r, y_r)
if isinstance(node.right, TreeNode):
self._splits_cnt += 1
self._build_tree(X_r, y_r, node.right, depth + 1)
def _traverse_tree(self, row: pd.Series) -> float:
node = self.tree
while isinstance(node, TreeNode):
if row[node.col_name] > node.split_value:
node = node.right
else:
node = node.left
return node.value
def print_tree(self) -> None:
print(self.tree)
def fit(self, X: pd.DataFrame, y: pd.Series) -> None:
self.tree = self._get_best_split(X, y)
self._build_tree(X, y, self.tree, depth=1)
def predict(self, X: pd.DataFrame) -> pd.Series:
y_pred = self.predict_proba(X)
flt = y_pred > 0.5
y_pred.loc[flt] = 1
y_pred.loc[~flt] = 0
return y_pred.astype(int)
def predict_proba(self, X: pd.DataFrame) -> pd.Series:
return X.apply(self._traverse_tree, axis=1)
if __name__ == '__main__':
df = pd.read_csv('https://archive.ics.uci.edu/static/public/267/banknote+authentication.zip', header=None)
df.columns = ['variance', 'skewness', 'kurtosis', 'entropy', 'target']
X, y = df.iloc[:, :4], df['target']
X_train = X.sample(int(len(X) * 0.8), random_state=41)
y_train = y.loc[X_train.index]
flt = ~X.index.isin(X_train.index)
y_test = y.loc[flt]
X_test = X.loc[flt]
t = MyTreeClf(max_depth=5, min_samples_split=5, max_leafs=20)
t.fit(X, y)
t.print_tree()
print("Prediction quality\n", pd.concat([t.predict(X_test).rename('pred'), y_test], axis=1).corr())