LeNet/softmax.py at master · afrozalm/LeNet · GitHub

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
import numpy as np
import time


class Softmax(object):
    def __init__(self, feature_dim,  classes=10):
        sigma = 2.0 / (feature_dim + classes)
        W = np.random.normal(loc=0.0,
                             scale=sigma,
                             size=(feature_dim, classes))
        b = np.random.normal(loc=0.001,
                             scale=0.0001,
                             size=(classes,))
        dE_dW = np.ndarray(W.shape)
        dE_db = np.ndarray(b.shape)
        self.params = [W, b]
        self.gradParams = [dE_dW, dE_db]
        self.feature_dim = feature_dim
        self.accumulated_gradParams = []
        self.acc_no = 0
        self.name = 'softmax'
        self.momentum_init = False
        self.i_t = 0

        def softmax(x):
            max_val = np.max(x)
            x = x - max_val
            return np.exp(x) / np.sum(np.exp(x), axis=0)

        def softmax_derivative(probs):
            return np.diag(probs) - np.dot(np.expand_dims(probs, 1),
                                           np.expand_dims(probs, 0))
        self.activation_fn = softmax
        self.activation_derivative = softmax_derivative

    def forward(self, feature_vec):
        # checking the last dim matches or not
        start_time = time.time()
        assert self.feature_dim == feature_vec.shape[-1]
        self.inp_vec = feature_vec

        W, b = self.params
        affine_transform = np.dot(feature_vec, W) + b
        probs = self.activation_fn(affine_transform)
        self.deriv_out = self.activation_derivative(probs)
        # print np.sum(W), np.sum(b)
        self.time_taken = time.time() - start_time
        return probs

    def accumulate_grads(self):
        self.acc_no += 1
        if self.accumulated_gradParams == []:
            self.accumulated_gradParams = map(lambda x: x * 1,
                                              self.gradParams)
        else:
            self.accumulated_gradParams = map(lambda x, y: x+y,
                                              self.accumulated_gradParams,
                                              self.gradParams)

    def backward(self, deltas):
        W = self.params[0]
        dE_dW, dE_db = self.gradParams

        gdY = self.deriv_out
        # print np.sum(gdY), np.sum(deltas)
        # print np.sum(dE_dW), 's+++', np.sum(dE_db)
        dE_dX = np.dot(np.dot(W, gdY), deltas)

        dE_dW = np.dot(np.expand_dims(self.inp_vec, axis=1),
                       np.expand_dims(np.dot(gdY, deltas), axis=0))

        dE_db = np.dot(gdY, deltas)
        # print np.sum(dE_dW), 's---', np.sum(dE_db)

        self.gradParams = [dE_dW, dE_db]
        self.accumulate_grads()
        return dE_dX

    def updateParams(self, hyperParams):
        alpha, beta1, beta2 = hyperParams
        epsilon = 10e-20
        self.i_t += 1

        self.gradParams = map(lambda x: x / self.acc_no,
                              self.accumulated_gradParams)
        self.acc_no = 0

        if not self.momentum_init:
            self.momentum_init = True
            self.momentum1 = []
            self.momentum2 = []
            for g in self.gradParams:
                self.momentum1.append(np.copy(g))
                self.momentum2.append(np.copy(np.square(g)))

        else:
            self.momentum1 = map(lambda m, g: np.add(beta1 * m,
                                                     (1 - beta1) * g),
                                 self.momentum1, self.gradParams)
            self.momentum2 = map(lambda m, g: np.add(beta2 * m,
                                                     (1 - beta2) *
                                                     np.square(g)),
                                 self.momentum2, self.gradParams)

        m_t = map(lambda x: x/(1 - beta1**self.i_t), self.momentum1)
        v_t = map(lambda x: x/(1 - beta2**self.i_t), self.momentum2)
        # print np.sum(v_t[0]), np.sum(v_t[1])

        self.params = map(lambda theta, m, v:
                          np.subtract(theta,
                                      alpha*np.divide(m,
                                                      np.sqrt(v + epsilon))),
                          self.params, m_t, v_t)

        self.accumulated_gradParams = []


if __name__ == '__main__':
    feature_dim = 5
    classes = 2
    s = Softmax(feature_dim=feature_dim, classes=classes)
    sample_input = np.ones((feature_dim,))
    print s.forward(sample_input)
    s.backward(np.ones((classes)))
    s.backward(np.ones((classes))*10e-5)
    s.updateParams([0.01, 0.9, 0.99])