File size: 5,051 Bytes
44466c7 525f2d6 44466c7 1bf893a 44466c7 525f2d6 1bf893a 44466c7 525f2d6 dc1e1b6 e4768e1 525f2d6 e4768e1 525f2d6 dc084c4 525f2d6 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 |
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.')
if st.session_state['female_bold'] and st.session_state['male_bold']:
st.subheader('Step 2: Generated 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']]
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)
print(male_generation)
st.session_state['male_continuations'] = [gen[0]['generated_text'].replace(prompt, '') for gen, prompt in
zip(male_generation, st.session_state['male_prompts'])]
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'])]
st.write('Generated {} male continuations'.format(len(st.session_state['male_continuations'])))
st.write('Generated {} female continuations'.format(len(st.session_state['female_continuations'])))
if st.session_state.get('male_continuations') and st.session_state.get('female_continuations'):
st.subheader('Step 3: Sample Generated Texts')
st.write('**Male Prompt:**', st.session_state['male_prompts'][0])
st.write('**Male Continuation:**', st.session_state['male_continuations'][0])
st.write('**Female Prompt:**', st.session_state['female_prompts'][0])
st.write('**Female Continuation:**', st.session_state['female_continuations'][0])
if st.button('Evaluate'):
st.subheader('Step 4: Regard Results')
regard = Regard("compare")
st.write('Computing regard results to compare male and female continuations...')
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)
st.write('Computing average regard results for comparative analysis...')
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)
|