写出from torchtext.legacy import data, 用RNN优化器的代码

import torch
from torchtext.legacy.data import TabularDataset, BucketIterator

# 定义数据集
train_data, valid_data, test_data = TabularDataset.splits(
    path='path/to/data',
    train='train.csv',
    validation='valid.csv',
    test='test.csv',
    format='csv',
    fields=[('text', text_field), ('label', label_field)]
)

# 定义迭代器
train_iter, valid_iter, test_iter = BucketIterator.splits(
    (train_data, valid_data, test_data),
    batch_sizes=(64, 64, 64),
    sort_key=lambda x: len(x.text),
    sort_within_batch=True,
    device=-1
)

# 定义模型
class RNNModel(nn.Module):
    def __init__(self, vocab_size, embedding_dim, hidden_dim, output_dim, n_layers, bidirectional, dropout):
        super().__init__()
        self.embedding = nn.Embedding(vocab_size, embedding_dim)
        self.rnn = nn.RNN(embedding_dim, hidden_dim, n_layers, bidirectional=bidirectional, dropout=dropout)
        self.fc = nn.Linear(hidden_dim*2, output_dim)
        self.dropout = nn.Dropout(dropout)
        
    def forward(self, x):
        # 将输入的形状从(batch_size, seq_len)转换为(batch_size, seq_len, embedding_dim)
        x = self.embedding(x)
        # 由于RNN的输入需要是(batch_size, seq_len, input_size)，而embedding后的形状为(batch_size, seq_len, embedding_dim)
        # 因此需要进行转置，将第二维和第三维调换位置
        x = x.permute(1, 0, 2)
        # 通过RNN网络获取隐藏状态
        output, hidden = self.rnn(x)
        # 将两个方向的隐藏状态拼接在一起
        hidden = self.dropout(torch.cat((hidden[-2,:,:], hidden[-1,:,:]), dim = 1))
        # 通过全连接层获取预测结果
        out = self.fc(hidden.squeeze(0))
        return out