model.py

from typing import Dict

import torch
import torch.nn as nn

device = "cpu"

class SeqClassifier(nn.Module):
    def __init__(
        self,
        embeddings: torch.tensor,
        hidden_size: int,
        num_layers: int,
        dropout: float,
        bidirectional: bool,
        num_class: int,
    ) -> None:
        super(SeqClassifier, self).__init__()
        self.embed = nn.Embedding.from_pretrained(embeddings, freeze=False)
        self.hidden_size=hidden_size
        self.num_layers=num_layers
        self.dropout=dropout
        self.bidirectional=bidirectional
        self.num_class=num_class

        # model architecture
        self.rnn = nn.LSTM(
            input_size=embeddings.size(1),
            hidden_size=hidden_size,
            num_layers=num_layers,
            dropout=dropout,
            bidirectional=bidirectional,
            batch_first=True
        )
        self.dropout_layer = nn.Dropout(p=self.dropout)
        self.fc = nn.Linear(self.encoder_output_size, num_class)

    @property
    def encoder_output_size(self) -> int:
        # calculate the output dimension of rnn
        if self.bidirectional:
            return self.hidden_size * 2
        else:
            return self.hidden_size

    def forward(self, batch) -> torch.Tensor:
        # 將輸入嵌入到詞嵌入空間，就是把詞索引換成詞向量
        embedded = self.embed(batch)

        # 過 LSTM 層
        rnn_output, _ = self.rnn(embedded)
        rnn_output = self.dropout_layer(rnn_output)

        if not self.training:
            last_hidden_state_forward = rnn_output[ -1, :self.hidden_size]  # 正向方向的隐藏状态
            last_hidden_state_backward = rnn_output[ 0, self.hidden_size:]  # 反向方向的隐藏状态
            combined_hidden_state = torch.cat((last_hidden_state_forward, last_hidden_state_backward), dim=0)

            # 通過全連接層
            logits = self.fc(combined_hidden_state)
            return logits  # 返回預測結果

        last_hidden_state_forward = rnn_output[:, -1, :self.hidden_size]  # 正向方向的隐藏状态
        last_hidden_state_backward = rnn_output[:, 0, self.hidden_size:]  # 反向方向的隐藏状态
        combined_hidden_state = torch.cat((last_hidden_state_forward, last_hidden_state_backward), dim=1)

        # 通過全連接層
        logits = self.fc(combined_hidden_state)
        return logits  # 返回預測結果