Explaining_Decision_of_Time_Series_Data/taylor.py at master · OpenXAIProject/Explaining_Decision_of_Time_Series_Data · 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
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
class Taylor:

    def __init__(self, activations, weights, conv_ksize, pool_ksize, conv_strides, pool_strides, name,part):

        self.last_ind = len(activations)
        for op in activations[::-1]:
            self.last_ind -= 1
            if any([word in op.name for word in ['conv', 'pooling', 'dense']]):
                break

        self.activations = activations
        self.weights = weights
        self.conv_ksize = conv_ksize
        self.pool_ksize = pool_ksize
        self.conv_strides = conv_strides
        self.pool_strides = pool_strides
        self.name = name
        self.part = part

    def __call__(self, logit):

        with tf.name_scope(self.name):
            Rs = []
            j = 0

            for i in range(len(self.activations) - 1):

                if i is self.last_ind:

                    if 'conv' in self.activations[i].name.lower():
                        Rs.append(self.backprop_conv_input(self.activations[i + 1], self.weights[j], Rs[-1], self.conv_strides))
                        #print ('backprop_conv_input: {},{}'.format(i,j))
                    else:
                        Rs.append(self.backprop_dense_input(self.activations[i + 1], self.weights[j], Rs[-1]))
                        #print ('backprop_dense_input: {},{}'.format(i,j))
                    continue

                if i is 0:
                    Rs.append(self.activations[i][:,logit,None])
                    Rs.append(self.backprop_dense(self.activations[i + 1], self.weights[j][:,logit,None], Rs[-1]))
                    #print ('backprop_dense: {},{}'.format(i,j))
                    j += 1
                    continue

                elif 'dense' in self.activations[i].name.lower():
                    Rs.append(self.backprop_dense(self.activations[i + 1], self.weights[j], Rs[-1]))
                    #print ('backprop_dense: {},{}'.format(i,j))
                    j += 1
                elif 'reshape' in self.activations[i].name.lower():
                    shape = self.activations[i + 1].get_shape().as_list()
                    #print ('reshape: {},{}'.format(i,j))
                    shape[0] = -1
                    Rs.append(tf.reshape(Rs[-1], shape))
                elif 'conv' in self.activations[i].name.lower():
                    Rs.append(self.backprop_conv(self.activations[i + 1], self.weights[j], Rs[-1], self.conv_strides))
                    #print ('backprop_conv: {},{}'.format(i,j))
                    j += 1
                elif 'pooling' in self.activations[i].name.lower():

                    # Apply average pooling backprop regardless of type of pooling layer used, following recommendations by Montavon et al.
                    # Uncomment code below if you want to apply the winner-take-all redistribution policy suggested by Bach et al.
                    #
                    if 'avg' in self.activations[i].name.lower():
                        pooling_type = 'avg'
                    else:
                        pooling_type = 'max'
                    Rs.append(self.backprop_pool(self.activations[i + 1], Rs[-1], self.pool_ksize, self.pool_strides, pooling_type))
                    #print ('backprop_pool: {},{}'.format(i,j))
                    #Rs.append(self.backprop_pool(self.activations[i + 1], Rs[-1], self.pool_ksize, self.pool_strides, 'avg'))
                else:
                    raise Error('Unknown operation.')
            if self.part =="whole":
                print ('whole LRP result')
                return Rs
            elif self.part =="part":
                print ('last input LRP result')
                return Rs[-1]

    def backprop_conv(self, activation, kernel, relevance, strides, padding='SAME'):
        W_p = tf.maximum(0., kernel)
        z = nn_ops.conv2d(activation, W_p, strides, padding) + 1e-10
        s = relevance / z
        c = nn_ops.conv2d_backprop_input(tf.shape(activation), W_p, s, strides, padding)

        return activation * c

    def backprop_pool(self, activation, relevance, ksize, strides, pooling_type, padding='SAME'):
        if pooling_type.lower() in 'avg':
            z = nn_ops.avg_pool(activation, ksize, strides, padding) + 1e-10
            s = relevance / z
            c = gen_nn_ops._avg_pool_grad(tf.shape(activation), s, ksize, strides, padding)
            #print ('--avg pooling')
            return activation * c
        else:
            z = nn_ops.max_pool(activation, ksize, strides, padding) + 1e-10
            s = relevance / z
            #c = gen_nn_ops._max_pool_grad(activation, z, s, ksize, strides, padding)
            c = gen_nn_ops.max_pool_grad(activation, z, s, ksize, strides, padding)
            #print ('--max pooling')
            return activation * c

    def backprop_dense(self, activation, kernel, relevance):
        W_p = tf.maximum(0., kernel)
        z = tf.matmul(activation, W_p) + 1e-10
        s = relevance / z
        c = tf.matmul(s, tf.transpose(W_p))
        return activation * c

    def backprop_conv_input(self, X, kernel, relevance, strides, padding='SAME', lowest=0., highest=1.):
        W_p = tf.maximum(0., kernel)
        W_n = tf.minimum(0., kernel)

        L = tf.ones_like(X, tf.float32) * lowest
        H = tf.ones_like(X, tf.float32) * highest

        z_o = nn_ops.conv2d(X, kernel, strides, padding)
        z_p = nn_ops.conv2d(L, W_p, strides, padding)
        z_n = nn_ops.conv2d(H, W_n, strides, padding)

        z = z_o - z_p - z_n + 1e-10
        s = relevance / z

        c_o = nn_ops.conv2d_backprop_input(tf.shape(X), kernel, s, strides, padding)
        c_p = nn_ops.conv2d_backprop_input(tf.shape(X), W_p, s, strides, padding)
        c_n = nn_ops.conv2d_backprop_input(tf.shape(X), W_n, s, strides, padding)

        return X * c_o - L * c_p - H * c_n

    #FIND THE ROOT POINT USING W^2 RULE
    def backprop_conv_input2(self, X, kernel, relevance, strides, padding='SAME', lowest=-1., highest=1.):
        V = kernel * kernel
        N =  V/ tf.ones_like(V , tf.float32) *1

        O = nn_ops.conv2d_backprop_input(tf.shape(X), N, relevance, strides, padding)

        return O

    #pixel intensity
    def backprop_dense_input(self, X, kernel, relevance, lowest=0., highest=1.):
        W_p = tf.maximum(0., kernel)
        W_n = tf.minimum(0., kernel)

        L = tf.ones_like(X, tf.float32) * lowest
        H = tf.ones_like(X, tf.float32) * highest

        z_o = tf.matmul(X, kernel)
        z_p = tf.matmul(L, W_p)
        z_n = tf.matmul(H, W_n)

        z = z_o - z_p - z_n + 1e-10
        s = relevance / z

        c_o = tf.matmul(s, tf.transpose(kernel))
        c_p = tf.matmul(s, tf.transpose(W_p))
        c_n = tf.matmul(s, tf.transpose(W_n))

        return X * c_o - L * c_p - H * c_n

    #FIND THE ROOT POINT USING W^2 RULE
    def backprop_dense_input2(self, X, kernel, relevance, lowest=0., highest=1.):
        V = kernel * kernel
        N =  V/ tf.ones_like(V , tf.float32) *1

        return nn_ops.conv2d_backprop_input(tf.shape(X), N, relevance, strides, padding)