gender-classification-python/convnet.py at main · yjh910/gender-classification-python · 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
import numpy as np
import pickle
from collections import OrderedDict
from layers import *

class ConvNet:
    def __init__(self, input_shape=(1,50,50), conv_num=1,
                 conv_param=[{'filter_num':30, 'filter_size':7, 'pad':0, 'stride':1, 'pool_size':2}],
                 hidden_size_list=([100]), output_size=2, weight_init_std=0.01):
        self.conv_num = conv_num
        self.input_shape = input_shape
        filter_num = []
        filter_size = []
        filter_pad = []
        filter_stride = []
        pool_size = []
        conv_output_size = []
        pool_output_size = []
        input_size = []
        for i in range(conv_num):
            filter_num.append(conv_param[i]['filter_num'])
            filter_size.append(conv_param[i]['filter_size'])
            filter_pad.append(conv_param[i]['pad'])
            filter_stride.append(conv_param[i]['stride'])
            pool_size.append(conv_param[i]['pool_size'])

        for i in range(conv_num-1):
            if i == 0:
                input_size.append(input_shape[1])
            conv_output_size.append((input_size[i] - filter_size[i] + 2*filter_pad[i]) / filter_stride[i] + 1)
            pool_output_size.append(conv_output_size[i]/pool_size[i])
            input_size.append(pool_output_size[i])
        conv_output_size.append((input_size[-1] - filter_size[-1] + 2*filter_pad[-1]) / filter_stride[-1] + 1)
        pool_output_size.append(int(filter_num[-1] * (conv_output_size[-1]/pool_size[-1]) * (conv_output_size[-1]/pool_size[-1])))

        params_num = len(hidden_size_list) + conv_num + 1
        self.params_num = params_num

        self.params = {}
        self.params['W1'] = np.random.randn(filter_num[0], input_shape[0],
                                            filter_size[0], filter_size[0]) * weight_init_std
        self.params['b1'] = np.zeros(filter_num[0])
        for idx in range(1, conv_num):
            self.params['W' + str(idx+1)] = np.random.randn(filter_num[idx], filter_num[idx-1],
                                                            filter_size[idx], filter_size[idx]) * weight_init_std
            self.params['b' + str(idx+1)] = np.zeros(filter_num[idx])

        self.params['W' + str(conv_num+1)] = np.random.randn(pool_output_size[conv_num-1], hidden_size_list[0])
        self.params['b' + str(conv_num+1)] = np.zeros(hidden_size_list[0])
        for idx in range(1, len(hidden_size_list)):
            self.params['W' + str(conv_num+1+idx)] = np.random.randn(hidden_size_list[idx-1], hidden_size_list[idx]) * weight_init_std
            self.params['b' + str(conv_num+1+idx)] = np.zeros(hidden_size_list[idx])
        self.params['W' + str(params_num)] = np.random.randn(hidden_size_list[-1], output_size) * weight_init_std
        self.params['b' + str(params_num)] = np.zeros(output_size)


        self.layers = OrderedDict()
        for idx in range(1, conv_num+1):
            self.layers['Conv' + str(idx)] = Convolution(self.params['W'+str(idx)], self.params['b'+str(idx)],
                                                         conv_param[idx-1]['stride'], conv_param[idx-1]['pad'])
            self.layers['Relu' + str(idx)] = Relu()
            self.layers['Pool' + str(idx)] = Pooling(pool_h=pool_size[idx-1], pool_w=pool_size[idx-1], stride=pool_size[idx-1])

        for idx in range(1, len(hidden_size_list)+1):
            idx_p = idx + conv_num
            self.layers['Affine' + str(idx)] = Affine(self.params['W'+str(idx_p)], self.params['b'+str(idx_p)])
            self.layers['Relu' + str(idx_p)] = Relu()
        self.layers['Affine' + str(len(hidden_size_list)+1)] = Affine(self.params['W'+str(params_num)], self.params['b'+str(params_num)])

        self.last_layer = SoftmaxWithLoss()

    def model(self):
        for idx in range(1, self.conv_num+1):
            print("Conv"+str(idx)+": ", end="")
            print(self.params['W'+str(idx)].shape)
        for idx in range(self.conv_num+1, self.params_num+1):
            print("Affine"+str(idx-self.conv_num)+": ", end="")
            print(self.params['W'+str(idx)].shape)

    def predict(self, x):
        for layer in self.layers.values():
            x = layer.forward(x)
        return x

    def loss(self, x, t):
        y = self.predict(x)
        return self.last_layer.forward(y, t)

    def accuracy(self, x, t):
        y = self.predict(x)
        y = np.argmax(y, axis=1)
        if t.ndim != 1 : t = np.argmax(t, axis=1)

        accuracy = np.sum(y == t) / float(x.shape[0])
        return accuracy

    def gradient(self, x, t):
        self.loss(x, t)

        dout = 1
        dout = self.last_layer.backward(dout)

        layers = list(self.layers.values())
        layers.reverse()
        for layer in layers:
            dout = layer.backward(dout)

        grads = {}
        for idx in range(1, self.conv_num+1):
            grads['W'+str(idx)] = self.layers['Conv'+str(idx)].dW
            grads['b'+str(idx)] = self.layers['Conv'+str(idx)].db
        for idx in range(self.conv_num+1, self.params_num+1):
            grads['W'+str(idx)] = self.layers['Affine'+str(idx-self.conv_num)].dW
            grads['b'+str(idx)] = self.layers['Affine'+str(idx-self.conv_num)].db

        return grads

    def save_params(self, file_name="face_params.pkl"):
        params = {}
        for key, val in self.params.items():
            params[key] = val
        with open(file_name, 'wb') as f:
            pickle.dump(params, f)
        print("params save complete!: {}".format(file_name))

    def load_params(self, file_name="face_params.pkl"):
        with open(file_name, 'rb') as f:
            params = pickle.load(f)
        for key, val in params.items():
            self.params[key] = val

        for idx in range(1, self.conv_num+1):
            self.layers['Conv'+str(idx)].W = self.params['W'+str(idx)]
            self.layers['Conv'+str(idx)].b = self.params['b'+str(idx)]

        for idx in range(self.conv_num+1, self.params_num+1):
            idx_p = idx - self.conv_num
            self.layers['Affine'+str(idx_p)].W = self.params['W'+str(idx)]
            self.layers['Affine'+str(idx_p)].b = self.params['b'+str(idx)]

        print("params load complete!: {}".format(file_name))