pages/1_Demo_1.py

import streamlit as st
import pandas as pd
from datasets import load_dataset
from random import sample
from utils.metric import Regard
from utils.model import gpt2
import os

# Set up the Streamlit interface
st.title('Gender Bias Analysis in Text Generation')


def check_password():
    def password_entered():
        if password_input == os.getenv('PASSWORD'):
            st.session_state['password_correct'] = True
        else:
            st.error("Incorrect Password, please try again.")

    password_input = st.text_input("Enter Password:", type="password")
    submit_button = st.button("Submit", on_click=password_entered)

    if submit_button and not st.session_state.get('password_correct', False):
        st.error("Please enter a valid password to access the demo.")


if not st.session_state.get('password_correct', False):
    check_password()
else:
    st.sidebar.success("Password Verified. Proceed with the demo.")

    if 'data_size' not in st.session_state:
        st.session_state['data_size'] = 10
    if 'bold' not in st.session_state:
        st.session_state['bold'] = load_dataset("AlexaAI/bold", split="train")
    if 'female_bold' not in st.session_state:
        st.session_state['female_bold'] = []
    if 'male_bold' not in st.session_state:
        st.session_state['male_bold'] = []

    st.subheader('Step 1: Set Data Size')
    data_size = st.slider('Select number of samples per category:', min_value=1, max_value=50,
                          value=st.session_state['data_size'])
    st.session_state['data_size'] = data_size

    if st.button('Show Data'):
        st.session_state['female_bold'] = sample(
            [p for p in st.session_state['bold'] if p['category'] == 'American_actresses'], data_size)
        st.session_state['male_bold'] = sample(
            [p for p in st.session_state['bold'] if p['category'] == 'American_actors'], data_size)

        st.write(f'Sampled {data_size} female and male American actors.')
        st.write('**Female Samples:**', pd.DataFrame(st.session_state['female_bold']))
        st.write('**Male Samples:**', pd.DataFrame(st.session_state['male_bold']))

    if st.session_state['female_bold'] and st.session_state['male_bold']:
        st.subheader('Step 2: Generate Text')

        if st.button('Generate Text'):
            GPT2 = gpt2()
            st.session_state['male_prompts'] = [p['prompts'][0] for p in st.session_state['male_bold']]
            st.session_state['female_prompts'] = [p['prompts'][0] for p in st.session_state['female_bold']]

            progress_bar = st.progress(0)

            st.write('Generating text for male prompts...')
            male_generation = GPT2.text_generation(st.session_state['male_prompts'], pad_token_id=50256, max_length=50,
                                                   do_sample=False, truncation=True)
            st.session_state['male_continuations'] = [gen[0]['generated_text'].replace(prompt, '') for gen, prompt in
                                                      zip(male_generation, st.session_state['male_prompts'])]

            progress_bar.progress(50)

            st.write('Generating text for female prompts...')
            female_generation = GPT2.text_generation(st.session_state['female_prompts'], pad_token_id=50256,
                                                     max_length=50, do_sample=False, truncation=True)
            st.session_state['female_continuations'] = [gen[0]['generated_text'].replace(prompt, '') for gen, prompt in
                                                        zip(female_generation, st.session_state['female_prompts'])]

            progress_bar.progress(100)
            st.write('Text generation completed.')

    if st.session_state.get('male_continuations') and st.session_state.get('female_continuations'):
        st.subheader('Step 3: Sample Generated Texts')

        st.write("Male Data Samples:")
        samples_df = pd.DataFrame({
            'Male Prompt': st.session_state['male_prompts'],
            'Male Continuation': st.session_state['male_continuations'],
        })
        st.write(samples_df)

        st.write("Female Data Samples:")
        samples_df = pd.DataFrame({
            'Female Prompt': st.session_state['female_prompts'],
            'Female Continuation': st.session_state['female_continuations']
        })
        st.write(samples_df)

        if st.button('Evaluate'):
            st.subheader('Step 4: Regard Results')
            regard = Regard("compare")
            st.write('Computing regard results to compare male and female continuations...')

            with st.spinner('Computing regard results...'):
                regard_results = regard.compute(data=st.session_state['male_continuations'],
                                                references=st.session_state['female_continuations'])
                st.write('**Raw Regard Results:**')
                st.json(regard_results)

                regard_results_avg = regard.compute(data=st.session_state['male_continuations'],
                                                    references=st.session_state['female_continuations'],
                                                    aggregation='average')
                st.write('**Average Regard Results:**')
                st.json(regard_results_avg)