File size: 6,089 Bytes
c77d957
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ffbf3ef
c77d957
 
 
 
 
 
 
 
 
 
 
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
import numpy as np

class LogReg():
    def __init__(self, max_iterations=100, threshold=0.5, learning_rate=0.01, class_weight=None, reg_eqn=None, reg_param=0, random_state=42):
        self.threshold = threshold
        self.max_iterations = max_iterations
        self.learning_rate = learning_rate
        self.class_weight = class_weight
        self.reg_eqn = reg_eqn
        self.reg_param=reg_param
        self.random_state=random_state
        self.weights = None
        self.bias = 0.0
        
    def __linear(self, x):
        return np.dot(x, self.weights) + self.bias
    
    def __sigmoid(self, z):
        z = np.clip(z, -500, 500)
        return 1.0 / (1.0 + np.exp(-z))
    
    def __loss(self, preds, y):
        m = len(y)
        eps = 1e-9
        preds = np.clip(preds, eps, 1 - eps)
        return - (1/m) * np.sum(y * np.log(preds) + (1 - y) * np.log(1 - preds))
    
    def fit(self, x, y):
        np.random.seed(42)
        num_samples, num_features = x.shape
        classes = np.unique(y)
        self.weights = np.random.randn(num_features) * 0.01
        self.bias = 0.0
        
        
        classes_weights = {}
        if self.class_weight is None:
            classes_weights = {label: 1.0 for label in classes}
        elif self.class_weight == 'balanced':
            for label in classes:
                classes_weights[label] = num_samples / (len(classes) * np.sum(y == label))
        else:
            classes_weights = self.class_weight
            
        loss_weights = np.ones_like(y, dtype=float)
        for label in classes:
            loss_weights[y == label] = classes_weights[label]
            
        weight_sum = np.sum(loss_weights)
            
        for i in range(self.max_iterations):
            linear_predictions = self.__linear(x)
            probas = self.__sigmoid(linear_predictions)
            cost = self.__loss(probas, y)
            
            if self.reg_eqn == 'L2':
                cost += self.reg_param * np.sum(self.weights ** 2)
            elif self.reg_eqn == 'L1':
                cost += self.reg_param * np.sum(np.abs(self.weights))
            
            #if i % 10 == 0:
                #print(f"Iteration {i}, Loss: {cost}")
            
            
            dw = np.dot(x.T, loss_weights*(probas-y)) / weight_sum
            db = np.sum(loss_weights*(probas-y)) / weight_sum
            
            if self.reg_eqn == 'L2':
                dw += (self.reg_param / weight_sum) * self.weights
            elif self.reg_eqn == 'L1':
                dw += (self.reg_param / weight_sum) * np.sign(self.weights)
            
            self.weights -= self.learning_rate * dw
            self.bias -= self.learning_rate * db
    
    def predict(self, test):
        linear_predictions = self.__linear(test)
        probas = self.__sigmoid(linear_predictions)
        return (probas >= self.threshold).astype(int)




class MultiLogReg():
    def __init__(self, max_iterations=100, learning_rate=0.01, class_weight=None, reg_eqn=None, reg_param=0, random_state=42):
        self.max_iterations = max_iterations
        self.learning_rate = learning_rate
        self.class_weight = class_weight
        self.reg_eqn = reg_eqn
        self.reg_param = reg_param
        self.random_state = random_state
        self.classes = None
        self.weights = None
        self.bias = None
        
    def __linear(self, x):
        return np.dot(x, self.weights) + self.bias
    
    def __softmax(self, z):
        e_z = np.exp(z - np.max(z, axis=-1, keepdims=True))
        return e_z / np.sum(e_z, axis=-1, keepdims=True)
    
    def __one_hot(self, y):
        n_samples = len(y)
        classes = np.unique(y)
        one_hot = np.zeros((n_samples, len(classes)))
        one_hot[np.arange(n_samples), y] = 1
        return one_hot
    
    def __loss(self, preds, y_one_hot):
        m = len(y_one_hot)
        eps = 1e-9
        preds = np.clip(preds, eps, 1 - eps)
        return -np.sum(y_one_hot * np.log(preds)) / m
    
    def fit(self, x, y):
        np.random.seed(self.random_state)
        num_samples, num_features = x.shape
        self.classes = np.unique(y)
        num_classes = len(self.classes)
        self.weights = np.random.randn(num_features, num_classes) * 0.01
        self.bias = np.zeros(num_classes)
        
        classes_weights = {}
        if self.class_weight is None:
            classes_weights = {label: 1.0 for label in self.classes}
        elif self.class_weight == 'balanced':
            for label in self.classes:
                classes_weights[label] = num_samples / (len(self.classes) * np.sum(y == label))
        else:
            classes_weights = self.class_weight
            
        loss_weights = np.ones_like(y, dtype=float)
        for label in self.classes:
            loss_weights[y == label] = classes_weights[label]
            
        weight_sum = np.sum(loss_weights)
        
        loss_weights = loss_weights.reshape(-1,1)
        y_one_hot = self.__one_hot(y)
            
        for i in range(self.max_iterations):
            linear_predictions = self.__linear(x)
            probas = self.__softmax(linear_predictions)
            cost = self.__loss(probas, y_one_hot)
            
            #if i % 10 == 0:
                #print(f"Iteration {i}, Loss: {cost}")
            
            dw = np.dot(x.T, loss_weights*(probas-y_one_hot)) / weight_sum
            db = np.sum(loss_weights*(probas-y_one_hot), axis=0) / weight_sum
            
            if self.reg_eqn == 'L2':
                dw += (self.reg_param / weight_sum) * self.weights
            elif self.reg_eqn == 'L1':
                dw += (self.reg_param / weight_sum) * np.sign(self.weights)
            
            self.weights -= self.learning_rate * dw
            self.bias -= self.learning_rate * db
    
    def predict(self, test):
        linear_predictions = self.__linear(test)
        probas = self.__softmax(linear_predictions)
        preds_indices = np.argmax(probas, axis=1)
        return self.classes[preds_indices]