PositionEncode.py

import math

import torch
import torch.nn as nn
import torch.nn.functional as F


class PositionalEncoding(nn.Module):

    def __init__(self, d_model, dropout=0.1, max_len=7000):
        super(PositionalEncoding, self).__init__()
        self.dropout = nn.Dropout(p=dropout)

        pe = torch.zeros(max_len, d_model)

        position = torch.arange(0, max_len).unsqueeze(1).float()
        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)
        self.register_buffer('pe', pe)

    def forward(self, input):
        """
        Args:
            input: N x T x D
        """
        length = input.size(1)
        tt = self.pe[:, :length]
        return tt


class PositionwiseFeedForward(nn.Module):

    def __init__(self, d_model, d_ff, dropout=0.1):
        super(PositionwiseFeedForward, self).__init__()
        self.w_1 = nn.Linear(d_model, d_ff)
        self.w_2 = nn.Linear(d_ff, d_model)
        self.dropout = nn.Dropout(dropout)
        self.layer_norm = nn.LayerNorm(d_model)

    def forward(self, x):
        residual = x
        output = self.w_2(F.relu(self.w_1(x)))
        output = self.dropout(output)
        output = self.layer_norm(output + residual)
        return output


# Another implementation
class PositionwiseFeedForwardUseConv(nn.Module):

    def __init__(self, d_in, d_hid, dropout=0.1):
        super(PositionwiseFeedForwardUseConv, self).__init__()
        self.w_1 = nn.Conv1d(d_in, d_hid, 1)  # position-wise
        self.w_2 = nn.Conv1d(d_hid, d_in, 1)  # position-wise
        self.layer_norm = nn.LayerNorm(d_in)
        self.dropout = nn.Dropout(dropout)

    def forward(self, x):
        residual = x
        output = x.transpose(2, 1)
        output = self.w_2(F.relu(self.w_1(output)))
        output = output.transpose(2, 1)
        output = self.dropout(output)
        output = self.layer_norm(output + residual)
        return output