Zekun Wu
update
d12a771
raw
history blame
16.9 kB
import time
import pandas as pd
import streamlit as st
import os
import json
from openai import AzureOpenAI
from model import create_models, configure_settings, load_documents_and_create_index, \
create_chat_prompt_template, execute_query
from datasets import Dataset, DatasetDict, load_dataset, concatenate_datasets
client = AzureOpenAI(azure_endpoint="https://personality-service.openai.azure.com/",
api_key=os.getenv("AZURE_OPENAI_KEY"), api_version="2024-02-15-preview")
TOKEN = os.getenv('hf_token')
def store_feedback(user_input, response, feedback, rating,repo):
dataset = load_dataset(repo, token=TOKEN, download_mode="force_redownload", ignore_verifications=True)
new_entry = pd.DataFrame({"user_input": [user_input], "response": [response], "feedback": [feedback], "rate": [rating]})
new_dataset = Dataset.from_pandas(new_entry)
updated_dataset = concatenate_datasets([dataset["train"], new_dataset])
updated_dataset.push_to_hub(repo, private=False, token=TOKEN)
# Function to generate a completion using OpenAI API
def generate_one_completion(message, temperature):
response = client.chat.completions.create(
model="personality_gpt4o",
temperature=temperature,
max_tokens=1000, # Adjust based on desired response length
frequency_penalty=0.2, # To avoid repetition
presence_penalty=0.2, # To introduce new topics
messages=message,
stream=False
)
return response
import json
def get_profile_str(profile):
bio_info = profile['bio_information']
main_profile = profile['main_profile']
red_flag = profile['red_flag']
motivation = profile['motivation']
profile_str = f"Bio Information:\n"
for key, value in bio_info.items():
profile_str += f"- {key.replace('_', ' ').title()}: {value}\n"
profile_str += f"\nMain Profile:\n"
for key, value in main_profile.items():
profile_str += f"- {key.title()}: {value['score']} - {value['summary']}\n"
profile_str += f"\nRed Flags:\n"
for key, value in red_flag.items():
profile_str += f"- {key.title()}: {value['score']} - {value['summary']}\n"
profile_str += f"\nMotivation:\n"
for key, value in motivation.items():
profile_str += f"- {key.title()}: {value['score']} - {value['summary']}\n"
return profile_str
def generate_prompt_from_profile(profile, version="TestTakersSummary"):
with open('prompts.json') as f:
prompt_sets = json.load(f)['Prompts']
prompt_templates = prompt_sets[version]
try:
# Fetching profile data
individual_name = profile['bio_information'].get('Name', 'the individual')
# Generating bio, profile, and red flags sections
bio_section = "\n".join(
[f"- {k.replace('_', ' ').title()}: {v}" for k, v in profile['bio_information'].items()])
main_profile_section = "\n".join(
[f"- {trait.title()}: {details['score']} - {details['summary']}" for trait, details in
profile['main_profile'].items()])
red_flags_section = "\n".join(
[f"- {trait.title()}: {details['score']} - {details['summary']}" for trait, details in
profile['red_flag'].items()])
motivation_section = "\n".join(
[f"- {trait.title()}: {details['score']} - {details['summary']}" for trait, details in
profile['motivation'].items()])
# Replacing placeholders in the prompts
prompts = [
x.replace('{{INDIVIDUAL_NAME}}', individual_name).replace('{{BIO}}', bio_section).replace('{{PROFILE}}',main_profile_section).replace(
'{{REDFLAGS_PROFILE}}', red_flags_section).replace('{{MOTIVATION_PROFILE}}', motivation_section) for x in prompt_templates]
# Compiling final prompt
prompt = "\n".join(prompts)
except KeyError as e:
return [{"role": "system", "content": f"Error processing profile data: missing {str(e)}"}]
message = [
{"role": "system", "content": prompt_sets[version][0]},
{"role": "user", "content": prompt}
]
return message
def display_profile_info(profile):
main_profile = profile["main_profile"]
red_flag = profile["red_flag"]
bio_info = profile["bio_information"]
st.sidebar.markdown("### Bio Information: ")
st.sidebar.markdown("\n".join([f"- **{key.replace('_', ' ')}**: {value}" for key, value in bio_info.items()]))
st.sidebar.markdown("### Main Profile: ")
st.sidebar.markdown("\n".join(
[f"- **{attribute}**: {details['score']} - {details['summary']}" for attribute, details in
main_profile.items()]))
st.sidebar.markdown("### Red Flags: ")
st.sidebar.markdown("\n".join(
[f"- **{attribute}**: {details['score']} - {details['summary']}" for attribute, details in red_flag.items()]))
st.sidebar.markdown("### Motivation: ")
st.sidebar.markdown("\n".join(
[f"- **{attribute}**: {details['score']} - {details['summary']}" for attribute, details in profile['motivation'].items()]))
def validate_json(profile):
required_keys = ['bio_information', 'main_profile', 'red_flag', 'motivation']
for key in required_keys:
if key not in profile:
return False, f"Key '{key}' is missing."
if not isinstance(profile[key], dict):
return False, f"'{key}' should be a dictionary."
return True, "JSON structure is valid."
def logout():
st.session_state['authenticated'] = False
st.session_state['profile'] = None
st.session_state['show_chat'] = None
st.session_state['analysis'] = None
st.rerun()
def main_app():
sidebar_components()
if st.button('Logout'):
logout()
# Streamlit app
st.title('Metaprofiling\'s Career Insight Analyzer Demo')
# Check if a profile is selected
if st.session_state['profile']:
profile = st.session_state['profile']
display_profile_info(profile) # Display the profile information
st.markdown("""
### Generation Temperature
Adjust the 'Generation Temperature' to control the creativity of the AI responses.
- A *lower temperature* (closer to 0.0) generates more predictable, conservative responses.
- A *higher temperature* (closer to 1.0) generates more creative, diverse responses.
""")
# Temperature slider
st.session_state['temperature'] = st.slider("", min_value=0.0, max_value=1.0, value=0.5, step=0.01)
# Allow user to choose from different versions of the prompt
st.session_state['version'] = st.selectbox("Select Prompt Version", ["TestTakersSummary", "ManagersSummary"])
# Generate and display prompt
if st.button(f'Analyze Profile ({st.session_state["version"]})'):
# with st.spinner('Generating completion...'):
prompt = generate_prompt_from_profile(profile, version=st.session_state['version'])
with st.chat_message("assistant"):
stream = client.chat.completions.create(
model="personality_gpt4o",
temperature=st.session_state['temperature'],
max_tokens=4096, # Adjust based on desired response length
frequency_penalty=0.2, # To avoid repetition
presence_penalty=0.2, # To introduce new topics
messages=prompt,
stream=True)
response = st.write_stream(stream)
# st.markdown(response_test_taker)
st.session_state['analysis'] = response
st.session_state['show_chat'] = True
st.rerun()
# display the response
if st.session_state['analysis']:
# Ask for feedback
st.markdown(st.session_state['analysis'])
# Ask the user to choose the type of feedback
feedback_type = st.selectbox(
"Select the type of feedback:",
["Report", "Coach"]
)
# Set the dataset identifier based on feedback type
if feedback_type == "Report":
dataset_id = "wu981526092/feedback_report"
else:
dataset_id = "wu981526092/feedback_coach"
st.markdown(f"Provide feedback on the {feedback_type.lower()}:")
criteria = {
"Faithfulness": "Are all claims made in the answer inferred from the given context, i.e., not hallucinated?",
"Answer Relevancy": "Is the answer relevant to the question?",
"Context Relevancy": "Is the context relevant to the question?",
"Correctness": "Is the answer factually correct, based on the context?",
"Clarity": "Is the answer explained clearly without the extensive jargon of the original document?",
"Completeness": "Is the question answered fully, with all parts and subquestions being addressed?",
}
ratings = {}
for criterion, description in criteria.items():
ratings[criterion] = st.slider(f"{criterion}: {description}", 0, 10, 5,key=f"{feedback_type} {criterion}")
feedback = st.text_input("Provide additional feedback on the response:",key=f"{feedback_type} feedback")
if st.button('Submit Report Feedback'):
if feedback_type == "Report":
store_feedback(str(generate_prompt_from_profile(profile, version=st.session_state['version'])), st.session_state['analysis'], feedback, str(ratings), dataset_id)
else:
store_feedback(str(st.session_state['coach_query']), str(st.session_state['coach_response']), feedback, str(ratings), dataset_id)
st.success("Feedback submitted successfully!")
else:
st.write("Please upload a profile JSON file or use the example profile.")
# Function to verify credentials and set the session state
def verify_credentials():
if st.session_state['username'] == os.getenv("username_app") and st.session_state['password'] == os.getenv(
"password_app"):
st.session_state['authenticated'] = True
else:
st.error("Invalid username or password")
# Login page
def login_page():
st.title("Welcome to Metaprofiling's Career Insight Analyzer Demo")
st.write(
"This application provides in-depth analysis and insights into professional profiles. Please log in to continue.")
# Description and Instructions
st.markdown("""
## How to Use This Application
- Enter your username and password in the sidebar.
- Click on 'Login' to access the application.
- Once logged in, you will be able to upload and analyze professional profiles.
""")
st.sidebar.write("Login:")
username = st.sidebar.text_input("Username") # , key='username')
password = st.sidebar.text_input("Password", type="password") # , key='password')
st.session_state['username'] = username
st.session_state['password'] = password
st.sidebar.button("Login", on_click=verify_credentials)
def sidebar_components():
with st.sidebar:
if st.button('Reset'):
st.session_state['profile'] = None
st.session_state['show_chat'] = None
st.session_state['analysis'] = None
st.rerun()
if not st.session_state['show_chat']:
# Instructions for JSON format
st.markdown("### JSON File Requirements:")
st.markdown("1. Must contain 'bio_information', 'main_profile', and 'red_flag' as top-level keys.")
st.markdown("2. Both keys should have dictionary values.")
st.markdown("### Choose the Definition:")
st.session_state['definition'] = st.selectbox("Select Definition", [1, 2, 3])
st.session_state['chat_context'] = st.selectbox("Select Chat Context", ["analysis", "profile"])
# File uploader
st.markdown("### Upload a profile JSON file")
uploaded_file = st.file_uploader("", type=['json'])
if uploaded_file is not None:
try:
profile_data = json.load(uploaded_file)
valid, message = validate_json(profile_data)
if valid:
st.session_state['profile'] = profile_data
else:
st.error(message)
except json.JSONDecodeError:
st.error("Invalid JSON file. Please upload a valid JSON file.")
# Button to load example profile
if st.button('Use Example Profile'):
if st.session_state['definition'] == 1:
file_name = "example_data_definition_1.json"
elif st.session_state['definition'] == 2:
file_name = "example_data_definition_2.json"
else:
file_name = "example_data_definition_3.json"
with open(file_name, 'r') as file:
st.session_state['profile'] = json.load(file)
else:
st.sidebar.title("Chat with Our Career Advisor")
#st.sidebar.markdown(
#"Hello, we hope you learned something about yourself in this report. This chat is here so you can ask any questions you have about your report! It’s also a great tool to get ideas about how you can use the information in your report for your personal development and achieving your current goals.")
# Name to be included in the questions
# name = st.session_state['profile']['bio_information'].get('Name', 'the individual')
# List of question templates where {} will be replaced with the name
question_templates = [
"What are the main risks associated with {}’s profile?",
"What are the implications of {}’s profile for working with others?",
# "What conclusions might we draw from his profile about {}’s style of leadership?",
# "Looking specifically at {}'s Red Flags, are there any particular areas of concern?",
# "Based on this profile, is {} better suited as a COO or a CEO?",
# "If speed of execution is important, based on his profile, how likely is {} to be able to achieve this?",
# "How is {} likely to react to business uncertainty and disruption?",
# "Based on his profile, what should a coaching plan designed for {} focus on?"
]
# Formatting each question template with the name
questions_list = [question.format("Test Taker") for question in question_templates]
# Prepare the questions for Markdown display
questions_markdown = "\n\n".join(
[f"Q{index + 1}: {question}" for index, question in enumerate(questions_list)])
# Code to display in the app
st.sidebar.markdown("### Suggest Questions")
st.sidebar.markdown(questions_markdown)
# st.sidebar.text_area("Suggested Questions", value=questions.choices[0].message.content, height=200, disabled=True)
user_input = st.sidebar.text_input("Ask a question about the profile analysis:")
llm, embed_model = create_models()
configure_settings(llm, embed_model)
index = load_documents_and_create_index()
if st.sidebar.button('Submit'):
if user_input:
st.session_state['coach_query'] = str(user_input)
if st.session_state['chat_context'] == "profile":
chat_prompt_template = create_chat_prompt_template(get_profile_str(st.session_state['profile']),st.session_state['definition'])
else:
chat_prompt_template = create_chat_prompt_template(st.session_state['analysis'],st.session_state['definition'])
st.session_state['coach_response'] = execute_query(index, chat_prompt_template, user_input)
st.sidebar.markdown(st.session_state['coach_response'])
# Initialize session state variables with default values if not already set
session_defaults = {
'show_chat': None,
'definition': 1,
'chat_context': "analysis",
'profile': None,
'analysis': None,
'temperature': 0,
'version': "",
'username': '',
'password': '',
'authenticated': False,
'coach_response':"",
'coach_query':""
}
for key, default in session_defaults.items():
if key not in st.session_state:
st.session_state[key] = default
# Show login or main app based on authentication status
if st.session_state['authenticated']:
main_app()
else:
login_page()