这里是基于encoder-decoder和注意力机制 模型代码，sigmoid对资产进行调仓。训练直接使用收益作为优化函数

import torch
import torch.nn as nn
from dataloader import create_dataloaders

class Attention(nn.Module):
    def __init__(self, hidden_dim):
        super(Attention, self).__init__()
        self.attn = nn.Linear(hidden_dim * 2, hidden_dim)
        self.v = nn.Parameter(torch.rand(hidden_dim))

    def forward(self, hidden, encoder_outputs):
        """
        参数:
        hidden (torch.Tensor): 解码器的隐藏状态，形状为 (batch_size, 1, hidden_dim)
        encoder_outputs (torch.Tensor): 编码器的输出序列，形状为 (batch_size, seq_len, hidden_dim)

        返回:
        attention_weights (torch.Tensor): 注意力权重，形状为 (batch_size, seq_len)
        """
        # 获取时间步数
        timestep = encoder_outputs.size(1)  # seq_len

        # 重复hidden，形状为 (batch_size, seq_len, hidden_dim)
        h = hidden.repeat(1, timestep, 1)  # h: (batch_size, seq_len, hidden_dim)

        # 拼接h和encoder_outputs，形状为 (batch_size, seq_len, hidden_dim * 2)
        energy = torch.tanh(self.attn(torch.cat((h, encoder_outputs), dim=2)))  # energy: (batch_size, seq_len, hidden_dim)

        # 转置energy以便进行批次矩阵乘法，形状为 (batch_size, hidden_dim, seq_len)
        energy = energy.transpose(1, 2)  # energy: (batch_size, hidden_dim, seq_len)

        # 重复v参数，形状为 (batch_size, 1, hidden_dim)
        v = self.v.repeat(encoder_outputs.size(0), 1).unsqueeze(1)  # v: (batch_size, 1, hidden_dim)