app.py

import re
import gradio as gr
from datasets import load_dataset
import torch
from torch.utils.data import random_split
from collections import Counter
import torch.nn as nn


class LSTMClassifier(nn.Module):
    def __init__(self, vocab_size, embedding_dim=200, hidden_dim=256):
        super(LSTMClassifier, self).__init__()
        self.embedding = nn.Embedding(vocab_size, embedding_dim, padding_idx=0)

        self.lstm = nn.LSTM(
            embedding_dim,
            hidden_dim,
            num_layers=2,
            batch_first=True,
            bidirectional=True,
            dropout=0.3,
        )

        # Dropout layer
        self.dropout = nn.Dropout(0.4)

        # Additional dense layers
        self.fc1 = nn.Linear(hidden_dim * 2, hidden_dim)
        self.fc2 = nn.Linear(hidden_dim, 2)

    def forward(self, x):
        embedded = self.embedding(x)

        lstm_out, (hidden, cell) = self.lstm(embedded)

        # Concatenate forward and backward hidden states
        hidden = torch.cat((hidden[-2, :, :], hidden[-1, :, :]), dim=1)
        hidden = self.dropout(hidden)

        # Additional layer with ReLU activation
        hidden = torch.relu(self.fc1(hidden))
        hidden = self.dropout(hidden)

        # Final classification layer
        out = self.fc2(hidden)
        return out


def create_vocabulary(ds, max_words=10000):
    word2idx = {
        "<PAD>": 0,
        "<UNK>": 1,
    }
    words = []
    for example in ds:
        text = example["sms"]
        text = text.lower()
        text = re.sub(r"[^\w\s]", "", text)
        words.extend(text.split())

    word_counts = Counter(words)
    common_words = word_counts.most_common(max_words - 2)
    for word, _ in common_words:
        word2idx[word] = len(word2idx)

    return word2idx


def create_splits(ds):
    # 80/20 split
    full_dataset = ds['train']
    train_size = int(0.8 * len(full_dataset))
    test_size = len(full_dataset) - train_size

    train_dataset, test_dataset = random_split(
        full_dataset,
        [train_size, test_size],
        generator=torch.Generator().manual_seed(42),
    )
    return train_dataset, test_dataset


ds = load_dataset("ucirvine/sms_spam")
train_dataset, test_dataset = create_splits(ds)
vocab = create_vocabulary(train_dataset)

# First recreate the model architecture
model = LSTMClassifier(len(vocab), 100)
# Load the saved state dict
model.load_state_dict(torch.load('best_model.pth'))
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

model = model.to(device)


def predict_text(model, text, word2idx, device, max_length=50):
    # Set model to evaluation mode
    model.eval()

    # Preprocess the text (same as training)
    text = text.lower()
    words = text.split()

    # Convert words to indices
    indices = [word2idx.get(word, word2idx['<UNK>']) for word in words]

    # Pad or truncate
    if len(indices) < max_length:
        indices += [word2idx['<PAD>']] * (max_length - len(indices))
    else:
        indices = indices[:max_length]

    # Convert to tensor
    with torch.no_grad():
        input_tensor = torch.tensor(indices).unsqueeze(
            0).to(device)  # Add batch dimension
        outputs = model(input_tensor)
        probabilities = torch.softmax(outputs, dim=1)
        prediction = torch.argmax(outputs, dim=1)

    return {
        'prediction': 'spam' if prediction.item() == 1 else 'ham',
        'confidence': probabilities[0][prediction].item()
    }


interface = gr.Interface(
    fn=lambda text: predict_text(model, text, vocab, device),
    inputs=gr.Textbox(lines=2, placeholder="Enter your text here..."),
    outputs=gr.Textbox(),
    title="SMS Spam Classifier",
    description="Enter a text message to predict if it's spam or ham.",
)

interface.launch(share=True)