pages/17_RNN.py

import streamlit as st
import torch
import torch.nn as nn
import torch.optim as optim
from torchtext.data.utils import get_tokenizer
from torchtext.vocab import build_vocab_from_iterator
from torchtext.datasets import IMDB
from torch.utils.data import DataLoader, random_split
import matplotlib.pyplot as plt
import seaborn as sns
import pandas as pd
import numpy as np
from collections import Counter
from torch.nn.utils.rnn import pad_sequence

# Define the RNN model
class RNN(nn.Module):
    def __init__(self, vocab_size, embed_size, hidden_size, output_size, n_layers, dropout):
        super(RNN, self).__init__()
        self.embedding = nn.Embedding(vocab_size, embed_size)
        self.rnn = nn.RNN(embed_size, hidden_size, n_layers, dropout=dropout, batch_first=True)
        self.fc = nn.Linear(hidden_size, output_size)
        self.dropout = nn.Dropout(dropout)

    def forward(self, x):
        x = self.dropout(self.embedding(x))
        h0 = torch.zeros(n_layers, x.size(0), hidden_size).to(device)
        out, _ = self.rnn(x, h0)
        out = self.fc(out[:, -1, :])
        return out

# Create a custom collate function to pad sequences
def collate_batch(batch):
    texts, labels = zip(*batch)
    text_lengths = [len(text) for text in texts]
    texts_padded = pad_sequence(texts, batch_first=True, padding_value=vocab["<pad>"])
    return texts_padded, torch.tensor(labels, dtype=torch.float), text_lengths

# Function to load the data
@st.cache_data
def load_data():
    tokenizer = get_tokenizer("basic_english")
    train_iter, test_iter = IMDB(split=('train', 'test'))

    def yield_tokens(data_iter):
        for _, text in data_iter:
            yield tokenizer(text)
    
    vocab = build_vocab_from_iterator(yield_tokens(train_iter), specials=["<unk>", "<pad>"])
    vocab.set_default_index(vocab["<unk>"])

    # Define the text and label processing pipelines
    text_pipeline = lambda x: vocab(tokenizer(x))
    label_pipeline = lambda x: 1 if x == 'pos' else 0

    # Process the data into tensors
    def process_data(data_iter):
        texts, labels = [], []
        for label, text in data_iter:
            texts.append(torch.tensor(text_pipeline(text), dtype=torch.long))
            labels.append(label_pipeline(label))
        return texts, torch.tensor(labels, dtype=torch.float)

    train_texts, train_labels = process_data(train_iter)
    test_texts, test_labels = process_data(test_iter)

    # Create DataLoaders
    train_dataset = list(zip(train_texts, train_labels))
    test_dataset = list(zip(test_texts, test_labels))

    train_size = int(0.8 * len(train_dataset))
    valid_size = len(train_dataset) - train_size
    train_dataset, valid_dataset = random_split(train_dataset, [train_size, valid_size])

    BATCH_SIZE = 64
    train_loader = DataLoader(train_dataset, batch_size=BATCH_SIZE, shuffle=True, collate_fn=collate_batch)
    valid_loader = DataLoader(valid_dataset, batch_size=BATCH_SIZE, shuffle=True, collate_fn=collate_batch)
    test_loader = DataLoader(test_dataset, batch_size=BATCH_SIZE, shuffle=True, collate_fn=collate_batch)

    return vocab, train_loader, valid_loader, test_loader

# Function to train the network
def train_network(net, iterator, optimizer, criterion, epochs):
    loss_values = []
    for epoch in range(epochs):
        epoch_loss = 0
        net.train()
        for texts, labels, _ in iterator:
            texts, labels = texts.to(device), labels.to(device)
            optimizer.zero_grad()
            predictions = net(texts).squeeze(1)
            loss = criterion(predictions, labels)
            loss.backward()
            optimizer.step()
            epoch_loss += loss.item()
        epoch_loss /= len(iterator)
        loss_values.append(epoch_loss)
        st.write(f'Epoch {epoch + 1}: loss {epoch_loss:.3f}')
    st.write('Finished Training')
    return loss_values

# Function to evaluate the network
def evaluate_network(net, iterator, criterion):
    epoch_loss = 0
    correct = 0
    total = 0
    all_labels = []
    all_predictions = []
    net.eval()
    with torch.no_grad():
        for texts, labels, _ in iterator:
            texts, labels = texts.to(device), labels.to(device)
            predictions = net(texts).squeeze(1)
            loss = criterion(predictions, labels)
            epoch_loss += loss.item()
            rounded_preds = torch.round(torch.sigmoid(predictions))
            correct += (rounded_preds == labels).sum().item()
            total += len(labels)
            all_labels.extend(labels.cpu().numpy())
            all_predictions.extend(rounded_preds.cpu().numpy())
    accuracy = 100 * correct / total
    st.write(f'Loss: {epoch_loss / len(iterator):.4f}, Accuracy: {accuracy:.2f}%')
    return accuracy, all_labels, all_predictions

# Load the data
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# Display a loading message with some vertical space
st.markdown("<div style='margin-top: 50px;'><b>Loading data...</b></div>", unsafe_allow_html=True)
vocab, train_loader, valid_loader, test_loader = load_data()

# Streamlit interface
st.title("RNN for Text Classification on IMDb Dataset")

st.write("""
This application demonstrates how to build and train a Recurrent Neural Network (RNN) for text classification using the IMDb dataset. You can adjust hyperparameters, visualize sample data, and see the model's performance.
""")

# Sidebar for input parameters
st.sidebar.header('Model Hyperparameters')
embed_size = st.sidebar.slider('Embedding Size', 50, 300, 100)
hidden_size = st.sidebar.slider('Hidden Size', 50, 300, 256)
n_layers = st.sidebar.slider('Number of RNN Layers', 1, 3, 2)
dropout = st.sidebar.slider('Dropout', 0.0, 0.5, 0.2, step=0.1)
learning_rate = st.sidebar.slider('Learning Rate', 0.001, 0.1, 0.01, step=0.001)
epochs = st.sidebar.slider('Epochs', 1, 20, 5)

# Create the network
vocab_size = len(vocab)
output_size = 1
net = RNN(vocab_size, embed_size, hidden_size, output_size, n_layers, dropout).to(device)
criterion = nn.BCEWithLogitsLoss()
optimizer = optim.Adam(net.parameters(), lr=learning_rate)

# Add vertical space
st.write('\n' * 10)

# Train the network
if st.sidebar.button('Train Network'):
    loss_values = train_network(net, train_loader, optimizer, criterion, epochs)
    
    # Plot the loss values
    plt.figure(figsize=(10, 5))
    plt.plot(range(1, epochs + 1), loss_values, marker='o')
    plt.title('Training Loss Over Epochs')
    plt.xlabel('Epoch')
    plt.ylabel('Loss')
    plt.grid(True)
    st.pyplot(plt)
    
    # Store the trained model in the session state
    st.session_state['trained_model'] = net

# Test the network
if 'trained_model' in st.session_state and st.sidebar.button('Test Network'):
    accuracy, all_labels, all_predictions = evaluate_network(st.session_state['trained_model'], test_loader, criterion)
    st.write(f'Test Accuracy: {accuracy:.2f}%')
    
    # Display results in a table
    st.write('Ground Truth vs Predicted')
    results = pd.DataFrame({
        'Ground Truth': all_labels,
        'Predicted': all_predictions
    })
    st.table(results.head(50))  # Display first 50 results for brevity

# Visualize some test results
def visualize_text_predictions(iterator, net):
    net.eval()
    samples = []
    with torch.no_grad():
        for texts, labels, _ in iterator:
            predictions = torch.round(torch.sigmoid(net(texts).squeeze(1)))
            samples.extend(zip(texts.cpu(), labels.cpu(), predictions.cpu()))
            if len(samples) >= 10:
                break
    return samples[:10]

if 'trained_model' in st.session_state and st.sidebar.button('Show Test Results'):
    samples = visualize_text_predictions(test_loader, st.session_state['trained_model'])
    st.write('Ground Truth vs Predicted for Sample Texts')
    for i, (text, true_label, predicted) in enumerate(samples):
        st.write(f'Sample {i+1}')
        st.text(' '.join([vocab.get_itos()[token] for token in text]))
        st.write(f'Ground Truth: {true_label.item()}, Predicted: {predicted.item()}')