model.py

import torch
import torch.nn as nn
from torch.utils.data import DataLoader, TensorDataset, random_split
import numpy as np

# Positional Encoding for time-awareness
class PositionalEncoding(nn.Module):
    def __init__(self, d_model, max_len=5000):
        super().__init__()
        pe = torch.zeros(max_len, d_model)
        position = torch.arange(0, max_len).unsqueeze(1)
        div_term = torch.exp(torch.arange(0, d_model, 2) * (-np.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, x):
        return x + self.pe[:, :x.size(1)]

# Transformer model for time series
class TransformerForecast(nn.Module):
    def __init__(self, input_dim, model_dim=64, nhead=4, num_layers=2, dropout=0.1, pred_len=6):
        super().__init__()
        self.input_dim = input_dim
        self.model_dim = model_dim
        self.pred_len = pred_len

        self.input_proj = nn.Linear(input_dim, model_dim)
        self.pos_encoder = PositionalEncoding(model_dim)

        encoder_layer = nn.TransformerEncoderLayer(d_model=model_dim, nhead=nhead, dropout=dropout, batch_first=True)
        self.transformer_encoder = nn.TransformerEncoder(encoder_layer, num_layers=num_layers)

        self.decoder = nn.Sequential(
            nn.Linear(model_dim, 128),
            nn.ReLU(),
            nn.Linear(128, input_dim * pred_len)
        )

    def forward(self, x):
        x = self.input_proj(x)
        x = self.pos_encoder(x)
        encoded = self.transformer_encoder(x)
        output = self.decoder(encoded[:, -1, :])
        return output.view(-1, self.pred_len, self.input_dim)