Zekun Wu commited on
Commit
c91dc45
β€’
1 Parent(s): 093cdcd
Files changed (1) hide show
  1. pages/1_Demo_1.py +77 -101
pages/1_Demo_1.py CHANGED
@@ -7,30 +7,9 @@ from utils.model import gpt2
7
  import os
8
 
9
  # Set up the Streamlit interface
10
- st.set_page_config(page_title="Gender Bias Analysis", page_icon="πŸ”", layout="wide")
11
  st.title('Gender Bias Analysis in Text Generation')
12
 
13
- # Initialize session state variables
14
- if 'password_correct' not in st.session_state:
15
- st.session_state['password_correct'] = False
16
- if 'data_size' not in st.session_state:
17
- st.session_state['data_size'] = 10
18
- if 'bold' not in st.session_state:
19
- st.session_state['bold'] = None
20
- if 'female_bold' not in st.session_state:
21
- st.session_state['female_bold'] = []
22
- if 'male_bold' not in st.session_state:
23
- st.session_state['male_bold'] = []
24
- if 'male_prompts' not in st.session_state:
25
- st.session_state['male_prompts'] = []
26
- if 'female_prompts' not in st.session_state:
27
- st.session_state['female_prompts'] = []
28
- if 'male_continuations' not in st.session_state:
29
- st.session_state['male_continuations'] = []
30
- if 'female_continuations' not in st.session_state:
31
- st.session_state['female_continuations'] = []
32
-
33
- # Password protection function
34
  def check_password():
35
  def password_entered():
36
  if password_input == os.getenv('PASSWORD'):
@@ -39,83 +18,25 @@ def check_password():
39
  st.error("Incorrect Password, please try again.")
40
 
41
  password_input = st.text_input("Enter Password:", type="password")
42
- st.button("Submit", on_click=password_entered)
43
 
44
- # Data loading function
45
- def load_data():
46
- if st.session_state['bold'] is None:
47
- st.session_state['bold'] = load_dataset("AlexaAI/bold", split="train")
48
 
49
- # Sampling function
50
- def sample_data(data_size):
51
- st.session_state['female_bold'] = sample(
52
- [p for p in st.session_state['bold'] if p['category'] == 'American_actresses'], data_size)
53
- st.session_state['male_bold'] = sample(
54
- [p for p in st.session_state['bold'] if p['category'] == 'American_actors'], data_size)
55
-
56
- # Text generation function
57
- def generate_text():
58
- GPT2 = gpt2()
59
- st.session_state['male_prompts'] = [p['prompts'][0] for p in st.session_state['male_bold']]
60
- st.session_state['female_prompts'] = [p['prompts'][0] for p in st.session_state['female_bold']]
61
-
62
- progress_bar = st.progress(0)
63
- st.write('Generating text for male prompts...')
64
- male_generation = GPT2.text_generation(st.session_state['male_prompts'], pad_token_id=50256, max_length=50,
65
- do_sample=False, truncation=True)
66
- st.session_state['male_continuations'] = [gen[0]['generated_text'].replace(prompt, '') for gen, prompt in
67
- zip(male_generation, st.session_state['male_prompts'])]
68
-
69
- progress_bar.progress(50)
70
-
71
- st.write('Generating text for female prompts...')
72
- female_generation = GPT2.text_generation(st.session_state['female_prompts'], pad_token_id=50256,
73
- max_length=50, do_sample=False, truncation=True)
74
- st.session_state['female_continuations'] = [gen[0]['generated_text'].replace(prompt, '') for gen, prompt in
75
- zip(female_generation, st.session_state['female_prompts'])]
76
-
77
- progress_bar.progress(100)
78
- st.write('Text generation completed.')
79
-
80
- # Display data samples function
81
- def display_samples():
82
- st.write("### Male Data Samples")
83
- samples_df = pd.DataFrame({
84
- 'Male Prompt': st.session_state['male_prompts'],
85
- 'Male Continuation': st.session_state['male_continuations'],
86
- })
87
- st.dataframe(samples_df)
88
-
89
- st.write("### Female Data Samples")
90
- samples_df = pd.DataFrame({
91
- 'Female Prompt': st.session_state['female_prompts'],
92
- 'Female Continuation': st.session_state['female_continuations']
93
- })
94
- st.dataframe(samples_df)
95
-
96
- # Evaluate regard function
97
- def evaluate_regard():
98
- regard = Regard("compare")
99
- st.write('Computing regard results to compare male and female continuations...')
100
-
101
- with st.spinner('Computing regard results...'):
102
- regard_results = regard.compute(data=st.session_state['male_continuations'],
103
- references=st.session_state['female_continuations'])
104
- st.write('**Raw Regard Results:**')
105
- st.json(regard_results)
106
-
107
- regard_results_avg = regard.compute(data=st.session_state['male_continuations'],
108
- references=st.session_state['female_continuations'],
109
- aggregation='average')
110
- st.write('**Average Regard Results:**')
111
- st.json(regard_results_avg)
112
-
113
- # Main app logic
114
- if not st.session_state['password_correct']:
115
  check_password()
116
  else:
117
  st.sidebar.success("Password Verified. Proceed with the demo.")
118
- load_data()
 
 
 
 
 
 
 
 
119
 
120
  st.subheader('Step 1: Set Data Size')
121
  data_size = st.slider('Select number of samples per category:', min_value=1, max_value=50,
@@ -123,17 +44,72 @@ else:
123
  st.session_state['data_size'] = data_size
124
 
125
  if st.button('Show Data'):
126
- sample_data(data_size)
 
 
 
 
127
  st.write(f'Sampled {data_size} female and male American actors.')
128
- display_samples()
 
129
 
130
  if st.session_state['female_bold'] and st.session_state['male_bold']:
131
  st.subheader('Step 2: Generate Text')
 
132
  if st.button('Generate Text'):
133
- generate_text()
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
134
 
135
- if st.session_state['male_continuations'] and st.session_state['female_continuations']:
136
- st.subheader('Step 3: Evaluate')
137
- display_samples()
138
  if st.button('Evaluate'):
139
- evaluate_regard()
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7
  import os
8
 
9
  # Set up the Streamlit interface
 
10
  st.title('Gender Bias Analysis in Text Generation')
11
 
12
+
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
13
  def check_password():
14
  def password_entered():
15
  if password_input == os.getenv('PASSWORD'):
 
18
  st.error("Incorrect Password, please try again.")
19
 
20
  password_input = st.text_input("Enter Password:", type="password")
21
+ submit_button = st.button("Submit", on_click=password_entered)
22
 
23
+ if submit_button and not st.session_state.get('password_correct', False):
24
+ st.error("Please enter a valid password to access the demo.")
 
 
25
 
26
+
27
+ if not st.session_state.get('password_correct', False):
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
28
  check_password()
29
  else:
30
  st.sidebar.success("Password Verified. Proceed with the demo.")
31
+
32
+ if 'data_size' not in st.session_state:
33
+ st.session_state['data_size'] = 10
34
+ if 'bold' not in st.session_state:
35
+ st.session_state['bold'] = load_dataset("AlexaAI/bold", split="train")
36
+ if 'female_bold' not in st.session_state:
37
+ st.session_state['female_bold'] = []
38
+ if 'male_bold' not in st.session_state:
39
+ st.session_state['male_bold'] = []
40
 
41
  st.subheader('Step 1: Set Data Size')
42
  data_size = st.slider('Select number of samples per category:', min_value=1, max_value=50,
 
44
  st.session_state['data_size'] = data_size
45
 
46
  if st.button('Show Data'):
47
+ st.session_state['female_bold'] = sample(
48
+ [p for p in st.session_state['bold'] if p['category'] == 'American_actresses'], data_size)
49
+ st.session_state['male_bold'] = sample(
50
+ [p for p in st.session_state['bold'] if p['category'] == 'American_actors'], data_size)
51
+
52
  st.write(f'Sampled {data_size} female and male American actors.')
53
+ st.write('**Female Samples:**', pd.DataFrame(st.session_state['female_bold']))
54
+ st.write('**Male Samples:**', pd.DataFrame(st.session_state['male_bold']))
55
 
56
  if st.session_state['female_bold'] and st.session_state['male_bold']:
57
  st.subheader('Step 2: Generate Text')
58
+
59
  if st.button('Generate Text'):
60
+ GPT2 = gpt2()
61
+ st.session_state['male_prompts'] = [p['prompts'][0] for p in st.session_state['male_bold']]
62
+ st.session_state['female_prompts'] = [p['prompts'][0] for p in st.session_state['female_bold']]
63
+
64
+ progress_bar = st.progress(0)
65
+
66
+ st.write('Generating text for male prompts...')
67
+ male_generation = GPT2.text_generation(st.session_state['male_prompts'], pad_token_id=50256, max_length=50,
68
+ do_sample=False, truncation=True)
69
+ st.session_state['male_continuations'] = [gen[0]['generated_text'].replace(prompt, '') for gen, prompt in
70
+ zip(male_generation, st.session_state['male_prompts'])]
71
+
72
+ progress_bar.progress(50)
73
+
74
+ st.write('Generating text for female prompts...')
75
+ female_generation = GPT2.text_generation(st.session_state['female_prompts'], pad_token_id=50256,
76
+ max_length=50, do_sample=False, truncation=True)
77
+ st.session_state['female_continuations'] = [gen[0]['generated_text'].replace(prompt, '') for gen, prompt in
78
+ zip(female_generation, st.session_state['female_prompts'])]
79
+
80
+ progress_bar.progress(100)
81
+ st.write('Text generation completed.')
82
+
83
+ if st.session_state.get('male_continuations') and st.session_state.get('female_continuations'):
84
+ st.subheader('Step 3: Sample Generated Texts')
85
+
86
+ st.write("Male Data Samples:")
87
+ samples_df = pd.DataFrame({
88
+ 'Male Prompt': st.session_state['male_prompts'],
89
+ 'Male Continuation': st.session_state['male_continuations'],
90
+ })
91
+ st.write(samples_df)
92
+
93
+ st.write("Female Data Samples:")
94
+ samples_df = pd.DataFrame({
95
+ 'Female Prompt': st.session_state['female_prompts'],
96
+ 'Female Continuation': st.session_state['female_continuations']
97
+ })
98
+ st.write(samples_df)
99
 
 
 
 
100
  if st.button('Evaluate'):
101
+ st.subheader('Step 4: Regard Results')
102
+ regard = Regard("compare")
103
+ st.write('Computing regard results to compare male and female continuations...')
104
+
105
+ with st.spinner('Computing regard results...'):
106
+ regard_results = regard.compute(data=st.session_state['male_continuations'],
107
+ references=st.session_state['female_continuations'])
108
+ st.write('**Raw Regard Results:**')
109
+ st.json(regard_results)
110
+
111
+ regard_results_avg = regard.compute(data=st.session_state['male_continuations'],
112
+ references=st.session_state['female_continuations'],
113
+ aggregation='average')
114
+ st.write('**Average Regard Results:**')
115
+ st.json(regard_results_avg)