import math
import torch
import torch.nn as nn

class PositionalEncoding(nn.Module):
    def __init__(self, d_model, max_len=5000):
        super(PositionalEncoding, self).__init__()

        # create a matrix of [max_len, d_model] filled with zeros
        pe = torch.zeros(max_len, d_model)

        # Create a column vector of positions [0, 1, 2, ..., max_len -1]
        position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)

        # Calculate the "division term" for the cos/sin math
        div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-math.log(10000.0) / d_model))

        # Fill even indices (0, 2, 4, ...) with Sine
        pe[:, 0::2] = torch.sin(position * div_term)

        # Fill odd indices (1, 3, 5, ...) with Cosine
        pe[:, 1::2] = torch.cos(position * div_term)

        # Add a batch dimension [1, max_len, d_model]
        pe = pe.unsqueeze(0)

        # register buffer ensures this is saved with the model but not trained
        self.register_buffer('pe', pe)

    def forward(self, x):
        # x is word embeddings [Batch, Seq_len, d-model]
        # We simply add the positional vectors to the word vectors
        x = x + self.pe[:, :x.size(1)]
        return x

class TransformerBlock(nn.Module):
    def __init__(self, embed_dim, num_heads, ff_dim, dropout=0.1):
        super(TransformerBlock, self).__init__()

        # Multihead Attention: How many different "perspectives" the model has (num_heads)
        self.attention = nn.MultiheadAttention(embed_dim, num_heads, batch_first=True)

        # Layer Normalization
        self.norm1 = nn.LayerNorm(embed_dim)
        self.norm2 = nn.LayerNorm(embed_dim)

        # Feed Forward Network
        self.ff = nn.Sequential(
            nn.Linear(embed_dim, ff_dim),
            nn.ReLU(),
            nn.Linear(ff_dim, embed_dim)
        )

        self.dropout = nn.Dropout(dropout)

    def forward(self, x):
        # x shape: [batch, seq_len, embed_dim]
        # Attention (Residual + Norm)
        attn_output, _ = self.attention(x, x, x)
        x = self.norm1(x + self.dropout(attn_output))

        # Feed Forward (Residual Connection +_Norm)
        ff_output = self.ff(x)
        x = self.norm2(x + self.dropout(ff_output))
        return x

class TransformerSentimentModel(nn.Module):
    def __init__(self, vocab_size, embed_dim, num_heads, ff_dim, num_layers, output_dim, max_len=300):
        super(TransformerSentimentModel, self).__init__()
        self.embedding = nn.Embedding(vocab_size, embed_dim)
        self.pos_encoding = PositionalEncoding(embed_dim, max_len)
        self.blocks = nn.ModuleList([
            TransformerBlock(embed_dim, num_heads, ff_dim) for _ in range(num_layers)
        ])

        self.dropout = nn.Dropout(0.5)
        self.fc = nn.Linear(embed_dim, output_dim)

    def forward(self, input_ids):
        # x: [batch, seq_len]
        x = self.embedding(input_ids) # [batch, seq_len, embed_dim]
        x = self.pos_encoding(x)

        for block in self.blocks:
            x = block(x)

        x = x[:, 0, :]
        return self.fc(self.dropout(x))