File size: 7,761 Bytes
8918ac7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers.activations import ACT2FN

class MaskedConv1d(nn.Conv1d):
    """A masked 1-dimensional convolution layer.



    Takes the same arguments as torch.nn.Conv1D, except that the padding is set automatically.



         Shape:

            Input: (N, L, in_channels)

            input_mask: (N, L, 1), optional

            Output: (N, L, out_channels)

    """

    def __init__(

        self,

        in_channels: int,

        out_channels: int,

        kernel_size: int,

        stride: int = 1,

        dilation: int = 1,

        groups: int = 1,

        bias: bool = True,

    ):
        """

        :param in_channels: input channels

        :param out_channels: output channels

        :param kernel_size: the kernel width

        :param stride: filter shift

        :param dilation: dilation factor

        :param groups: perform depth-wise convolutions

        :param bias: adds learnable bias to output

        """
        padding = dilation * (kernel_size - 1) // 2
        super().__init__(
            in_channels,
            out_channels,
            kernel_size,
            stride=stride,
            dilation=dilation,
            groups=groups,
            bias=bias,
            padding=padding,
        )

    def forward(self, x, input_mask=None):
        if input_mask is not None:
            x = x * input_mask
        return super().forward(x.transpose(1, 2)).transpose(1, 2)


class Attention1dPooling(nn.Module):
    def __init__(self, hidden_size):
        super().__init__()
        self.layer = MaskedConv1d(hidden_size, 1, 1)

    def forward(self, x, input_mask=None):
        batch_szie = x.shape[0]
        attn = self.layer(x)
        attn = attn.view(batch_szie, -1)
        if input_mask is not None:
            attn = attn.masked_fill_(
                ~input_mask.view(batch_szie, -1).bool(), float("-inf")
            )
        attn = F.softmax(attn, dim=-1).view(batch_szie, -1, 1)
        out = (attn * x).sum(dim=1)
        return out

class Attention1dPoolingProjection(nn.Module):
    def __init__(self, hidden_size, num_labels, dropout=0.25) -> None:
        super(Attention1dPoolingProjection, self).__init__()
        self.linear = nn.Linear(hidden_size, hidden_size)
        self.dropout = nn.Dropout(dropout)
        self.relu = nn.ReLU()
        self.final = nn.Linear(hidden_size, num_labels)

    def forward(self, x):
        x = self.linear(x)
        x = self.dropout(x)
        x = self.relu(x)
        x = self.final(x)
        return x

class Attention1dPoolingHead(nn.Module):
    """Outputs of the model with the attention1d"""

    def __init__(

        self, hidden_size: int, num_labels: int, dropout: float = 0.25

    ):  # [batch x sequence(751) x embedding (1280)] --> [batch x embedding] --> [batch x 1]
        super(Attention1dPoolingHead, self).__init__()
        self.attention1d = Attention1dPooling(hidden_size)
        self.attention1d_projection = Attention1dPoolingProjection(hidden_size, num_labels, dropout)

    def forward(self, x, input_mask=None):
        x = self.attention1d(x, input_mask=input_mask.unsqueeze(-1))
        x = self.attention1d_projection(x)
        return x

class MeanPooling(nn.Module):
    """Mean Pooling for sentence-level classification tasks."""

    def __init__(self):
        super().__init__()

    def forward(self, features, input_mask=None):
        if input_mask is not None:
            # Applying input_mask to zero out masked values
            masked_features = features * input_mask.unsqueeze(2)
            sum_features = torch.sum(masked_features, dim=1)
            mean_pooled_features = sum_features / input_mask.sum(dim=1, keepdim=True)
        else:
            mean_pooled_features = torch.mean(features, dim=1)
        return mean_pooled_features


class MeanPoolingProjection(nn.Module):
    """Mean Pooling with a projection layer for sentence-level classification tasks."""

    def __init__(self, hidden_size, num_labels, dropout=0.25):
        super().__init__()
        self.dense = nn.Linear(hidden_size, hidden_size)
        self.dropout = nn.Dropout(dropout)
        self.out_proj = nn.Linear(hidden_size, num_labels)

    def forward(self, mean_pooled_features):
        x = self.dropout(mean_pooled_features)
        x = self.dense(x)
        x = ACT2FN['gelu'](x)
        x = self.dropout(x)
        x = self.out_proj(x)
        return x


class MeanPoolingHead(nn.Module):
    """Mean Pooling Head for sentence-level classification tasks."""

    def __init__(self, hidden_size, num_labels, dropout=0.25):
        super().__init__()
        self.mean_pooling = MeanPooling()
        self.mean_pooling_projection = MeanPoolingProjection(hidden_size, num_labels, dropout)

    def forward(self, features, input_mask=None):
        mean_pooling_features = self.mean_pooling(features, input_mask=input_mask)
        x = self.mean_pooling_projection(mean_pooling_features)
        return x


class LightAttentionPoolingHead(nn.Module):
    def __init__(self, hidden_size=1280, num_labels=11, dropout=0.25, kernel_size=9, conv_dropout: float = 0.25):
        super(LightAttentionPoolingHead, self).__init__()

        self.feature_convolution = nn.Conv1d(hidden_size, hidden_size, kernel_size, stride=1,
                                             padding=kernel_size // 2)
        self.attention_convolution = nn.Conv1d(hidden_size, hidden_size, kernel_size, stride=1,
                                               padding=kernel_size // 2)

        self.softmax = nn.Softmax(dim=-1)

        self.dropout = nn.Dropout(conv_dropout)

        self.linear = nn.Sequential(
            nn.Linear(2 * hidden_size, 32),
            nn.Dropout(dropout),
            nn.ReLU(),
            nn.BatchNorm1d(32)
        )

        self.output = nn.Linear(32, num_labels)

    def forward(self, x: torch.Tensor, mask, **kwargs) -> torch.Tensor:
        """

        Args:

            x: [batch_size, sequence_length, hidden_size] embedding tensor that should be classified

            mask: [batch_size, sequence_length] mask corresponding to the zero padding used for the shorter sequecnes in the batch. All values corresponding to padding are False and the rest is True.



        Returns:

            classification: [batch_size,num_labels] tensor with logits

        """
        x = x.permute(0, 2, 1)  # [batch_size, hidden_size, sequence_length]
        o = self.feature_convolution(x)  # [batch_size, hidden_size, sequence_length]
        o = self.dropout(o)  # [batch_gsize, hidden_size, sequence_length]
        attention = self.attention_convolution(x)  # [batch_size, hidden_size, sequence_length]

        # mask out the padding to which we do not want to pay any attention (we have the padding because the sequences have different lenghts).
        # This padding is added by the dataloader when using the padded_permuted_collate function in utils/general.py
        attention = attention.masked_fill(mask[:, None, :] == False, -1e9)

        # code used for extracting embeddings for UMAP visualizations
        # extraction =  torch.sum(x * self.softmax(attention), dim=-1)
        # extraction = self.id0(extraction)

        o1 = torch.sum(o * self.softmax(attention), dim=-1)  # [batchsize, hidden_size]
        o2, _ = torch.max(o, dim=-1)  # [batchsize, hidden_size]
        o = torch.cat([o1, o2], dim=-1)  # [batchsize, 2*hidden_size]
        o = self.linear(o)  # [batchsize, 32]
        return self.output(o)  # [batchsize, num_labels]