SparseTransformer/modules.py at main · Lin956/SparseTransformer · 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
import torch
import torch.nn as nn
from Layers import MultiHeadAttention, CrossAttention
import torch.nn.functional as F
import math


class Encoder(nn.Module):
    def __init__(self, en_num_layers, embed_size, heads, map_dim):
        super(Encoder, self).__init__()
        self.encoderlayer = EncoderLayer(embed_size, heads, map_dim)
        self.en_num_layers = en_num_layers

    def forward(self, input):
        for i in range(self.en_num_layers):
            out = self.encoderlayer(input)
            input = out
        return out


# input: [C, T, embed_size]
# ouput: [C, T, embed_size]
class EncoderLayer(nn.Module):
    def __init__(self, embed_size, heads, map_dim):
        super(EncoderLayer, self).__init__()
        self.attn = MultiHeadAttention(embed_size, heads)
        self.fc1 = nn.Linear(embed_size, map_dim)
        self.fc2 = nn.Linear(map_dim, embed_size)
        self.norm1 = nn.LayerNorm(embed_size)
        self.norm2 = nn.LayerNorm(embed_size)

    def forward(self, input):
        x = self.attn(input)
        x = self.norm1(x + input)
        out = self.fc2(F.relu(self.fc1(x)))
        out = self.norm2(x + out)
        return out


# x_en: [C, T, embed_size]
# y_de: [C, out_T_dim, embed_size]
class Decoder(nn.Module):
    def __init__(self, dec_num_layers, embed_size, heads, T, out_T_dim, map_dim):
        super(Decoder, self).__init__()
        self.dec_num_layers = dec_num_layers
        self.decoderlayer = DecoderLayer(embed_size, heads, T, out_T_dim, map_dim)

    def forward(self, x_en, y_de):
        for i in range(self.dec_num_layers):
            out = self.decoderlayer(x_en, y_de)
            y_de = out
        return out


# out: [C, out_T_dim, embed_size]
class DecoderLayer(nn.Module):
    def __init__(self, embed_size, heads, T, out_T_dim, map_dim):
        super(DecoderLayer, self).__init__()
        self.attn = MultiHeadAttention(embed_size, heads)
        self.crossattn = CrossAttention(embed_size, heads, T, out_T_dim)
        self.fc1 = nn.Linear(embed_size, map_dim)
        self.fc2 = nn.Linear(map_dim, embed_size)
        self.norm1 = nn.LayerNorm(embed_size)
        self.norm2 = nn.LayerNorm(embed_size)
        self.norm3 = nn.LayerNorm(embed_size)

    def forward(self, x_en, y_de):
        y_de = self.norm1(self.attn(y_de) + y_de)
        crossout = self.norm2(self.crossattn(x_en, y_de) + y_de)
        out = self.fc2(F.relu(self.fc1(crossout)))
        return self.norm3(out + crossout)


"""
x = torch.rand((1, 20, 512))
y = torch.rand((1, 1, 512))
model = Decoder(6, 512, 8, 20, 1, 2048)
out = model(x, y)
print(out.shape)
"""


# input: [C, T, embed_size]
# output: [C, T, embed_size]
class PositionalEncoding(nn.Module):
    def __init__(self, d_model, dropout=0.1, max_len=5000):
        super(PositionalEncoding, self).__init__()

        pe = torch.zeros(max_len, d_model)
        position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)

        div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-math.log(10000.0) / d_model))

        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)

        pe = pe.unsqueeze(0).transpose(0, 1)

        self.register_buffer('pe', pe)

    def forward(self, x):
        x = x + self.pe[:x.size(1), :].squeeze(1)
        return x


"""
x = torch.rand((1, 3, 1))
model = PositionalEncoding(512)
y = model(x)
print(y.shape)
"""


class discriminator(nn.Module):
    def __init__(self, embed_size, out_T_dim):
        super(discriminator, self).__init__()
        self.fc1 = nn.Linear(1, embed_size)
        self.fc2 = nn.Linear(embed_size, 2048)
        self.pe = PositionalEncoding(embed_size)
        self.fc3 = nn.Linear(2048, 1024)
        self.fc4 = nn.Linear(1024, out_T_dim)

    def forward(self, input):
        input.unsqueeze_(0)  # [1, C, T]
        x = input.permute(1, 2, 0)  # [C, T, 1]
        x_pe = self.pe(x)
        x = F.relu(self.fc1(x)) + x_pe  # [C, T, embed_size]

        x = F.leaky_relu(self.fc2(x))
        x = F.leaky_relu(self.fc3(x))
        out = F.leaky_relu(self.fc4(x))
        out.squeeze_(2)

        return out