File size: 5,062 Bytes
ab81f90
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
166
167
168
169
170
import cupy as cp
from utils.im2col import im2col_indices, col2im_indices

class Conv2D:
    def __init__(self, in_channels, out_channels, kernel_size, stride=1, padding=0):
        self.cache = {}
        self.in_channels = in_channels
        self.out_channels = out_channels
        self.kernel_size = kernel_size
        self.stride = stride
        self.padding = padding
        
        fan_in = in_channels * kernel_size * kernel_size
        self.weights = cp.random.randn(out_channels, in_channels, kernel_size, kernel_size) * cp.sqrt(2.0 / fan_in)
        self.biases = cp.zeros((out_channels, 1))

    def forward(self, X):
        n_samples, _, h_in, w_in = X.shape
        
        h_out = (h_in - self.kernel_size + 2 * self.padding) // self.stride + 1
        w_out = (w_in - self.kernel_size + 2 * self.padding) // self.stride + 1
        
        X_col = im2col_indices(X, self.kernel_size, self.kernel_size, self.padding, self.stride)
        
        W_row = self.weights.reshape(self.out_channels, -1)
        
        Z_col = W_row @ X_col + self.biases
        
        Z = Z_col.reshape(self.out_channels, h_out, w_out, n_samples)
        Z = Z.transpose(3, 0, 1, 2)
        
        self.cache['X'] = X
        self.cache['X_col'] = X_col
        self.cache['W_row'] = W_row
        
        return Z

    def backward(self, dZ):
        X = self.cache['X']
        X_col = self.cache['X_col']
        W_row = self.cache['W_row']
        
        dZ_col = dZ.transpose(1, 2, 3, 0).reshape(self.out_channels, -1)
        
        dW = dZ_col @ X_col.T
        db = cp.sum(dZ_col, axis=1, keepdims=True)
        
        dX_col = W_row.T @ dZ_col
        
        dX = col2im_indices(dX_col, X.shape, self.kernel_size, self.kernel_size, self.padding, self.stride)
        
        self.dW = dW.reshape(self.weights.shape)
        self.db = db
        
        return dX

class MaxPool2D:
    def __init__(self, pool_size, stride=None):
        self.cache = {}
        self.pool_size = pool_size
        self.stride = stride if stride is not None else pool_size
        
    def forward(self, X):
        N, C, H, W = X.shape
        
        out_h = (H - self.pool_size) // self.stride + 1
        out_w = (W - self.pool_size) // self.stride + 1
        
        X_reshaped = X.reshape(N * C, 1, H, W)
        
        X_col = im2col_indices(X_reshaped, self.pool_size, self.pool_size, padding=0, stride=self.stride)
        
        max_idx = cp.argmax(X_col, axis=0)
        out = X_col[max_idx, cp.arange(max_idx.size)]
        
        out = out.reshape(out_h, out_w, N, C).transpose(2, 3, 0, 1)
        
        self.cache['X_shape'] = X.shape
        self.cache['X_col_shape'] = X_col.shape
        self.cache['max_idx'] = max_idx
        
        return out
    
    def backward(self, dZ):
        X_shape = self.cache['X_shape']
        X_col_shape = self.cache['X_col_shape']
        max_idx = self.cache['max_idx']
        N, C, H, W = X_shape
        
        dZ_flat = dZ.transpose(2, 3, 0, 1).ravel()
        
        dX_col = cp.zeros(X_col_shape)
        
        dX_col[max_idx, cp.arange(max_idx.size)] = dZ_flat
        
        dX_reshaped = col2im_indices(dX_col, (N * C, 1, H, W), self.pool_size, self.pool_size, padding=0, stride=self.stride)
        
        dX = dX_reshaped.reshape(X_shape)
        
        return dX

class Flatten:
    def __init__(self):
        self.cache = {}
        
    def forward(self, X):
        self.cache['X_shape'] = X.shape
        batch_size = X.shape[0]
        return X.reshape(batch_size, -1)
        
    def backward(self, dZ):
        return dZ.reshape(self.cache['X_shape'])
    
class Dropout:
    def __init__(self, p=0.5):
        self.cache = {}
        self.p = p
        self.training = True

    def forward(self, X):
        if not self.training:
            return X
            
        self.mask = (cp.random.rand(*X.shape) > self.p) / (1.0 - self.p)
        return X * self.mask

    def backward(self, dZ):
        return dZ * self.mask

class Linear:
    def __init__(self, input_dimension, output_dimension):
        self.cache = {}
        
        self.input_dimension = input_dimension
        self.output_dimension = output_dimension
        
        self.weights = cp.random.randn(input_dimension, output_dimension) * cp.sqrt(2.0 / input_dimension)
        self.biases = cp.zeros((1, output_dimension))
    
    def forward(self, X):
        Z = X @ self.weights + self.biases
        
        self.cache['X'] = X
        
        return Z
     
    def backward(self, dZ):
        X = self.cache['X']
        
        dW = X.T @ dZ
        db = cp.sum(dZ, axis=0, keepdims=True)
        dX = dZ @ self.weights.T
        
        self.dW = dW
        self.db = db
        
        return dX
    
class ReLU:
    def __init__(self):
        self.cache = {}
    
    def forward(self, X):
        self.cache['X'] = X
        return cp.maximum(0, X)
    
    def backward(self, dA):
        X = self.cache['X']
        dX = dA * (X > 0) 
        return dX