app.py

import streamlit as st
import pandas as pd
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

# Define the available models to choose from
models = {
    'BERT': 'bert-base-uncased',
    'RoBERTa': 'roberta-base',
    'DistilBERT': 'distilbert-base-uncased'
}

# Create a drop-down menu to select the model
model_name = st.sidebar.selectbox('Select Model', list(models.keys()))

# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained(models[model_name])
model = AutoModelForSequenceClassification.from_pretrained(models[model_name])

# Define the classes and their corresponding labels
classes = {
    0: 'Non-Toxic',
    1: 'Toxic',
    2: 'Severely Toxic',
    3: 'Obscene',
    4: 'Threat',
    5: 'Insult',
    6: 'Identity Hate'
}

# Create a function to generate the toxicity predictions
@st.cache(allow_output_mutation=True)
def predict_toxicity(tweet, model, tokenizer):
    # Preprocess the text
    inputs = tokenizer(tweet, padding=True, truncation=True, return_tensors='pt')
    # Get the predictions from the model
    outputs = model(**inputs)
    predictions = torch.nn.functional.softmax(outputs.logits, dim=1).detach().numpy()
    # Get the class with the highest probability
    predicted_class = int(predictions.argmax())
    predicted_class_label = classes[predicted_class]
    predicted_prob = predictions[0][predicted_class]
    return predicted_class_label, predicted_prob

# Create a table to display the toxicity predictions
def create_table(predictions):
    data = {'Tweet': [], 'Highest Toxicity Class': [], 'Probability': []}
    for tweet, prediction in predictions.items():
        data['Tweet'].append(tweet)
        data['Highest Toxicity Class'].append(prediction[0])
        data['Probability'].append(prediction[1])
    df = pd.DataFrame(data)
    return df

# Create the user interface
st.title('Toxicity Prediction App')
tweet_input = st.text_input('Enter a tweet:')
if st.button('Predict'):
    # Generate the toxicity prediction for the tweet using the selected model
    predicted_class_label, predicted_prob = predict_toxicity(tweet_input, model, tokenizer)
    prediction_text = f'Prediction: {predicted_class_label} ({predicted_prob:.2f})'
    st.write(prediction_text)

    # Display the toxicity predictions in a table
    predictions = {tweet_input: (predicted_class_label, predicted_prob)}
    table = create_table(predictions)
    st.table(table)