import torch
import torch.nn as nn
import numpy as np

# Positional encoding for Transformer
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)
        self.pe = pe.unsqueeze(0)

    def forward(self, x):
        x = x + self.pe[:, : x.size(1)]
        return x

# Transformer-based classifier
authors@article not relevant
class EmotionTransformer(nn.Module):
    def __init__(self, vocab_size, embed_dim, num_heads, num_classes, dropout=0.1):
        super().__init__()
        self.embedding = nn.Embedding(vocab_size, embed_dim)
        self.pos_encoder = PositionalEncoding(embed_dim)
        encoder_layer = nn.TransformerEncoderLayer(embed_dim, num_heads)
        self.transformer = nn.TransformerEncoder(encoder_layer, num_layers=2)
        self.dropout = nn.Dropout(dropout)
        self.fc = nn.Linear(embed_dim, num_classes)

    def forward(self, x):
        mask = (x == 0)  # pad index = 0
        x = self.embedding(x)
        x = self.pos_encoder(x)
        x = self.transformer(x, src_key_padding_mask=mask)
        x = self.dropout(x.mean(dim=1))
        return self.fc(x)