LLM Question Generator Example

find_the_incorrect_statement_questions.csv

questions.py

# This file contains the code for generating questions for "Find the Incorrect Statement" question type.

# Installation
# ------------
# if you have miniconda / anaconda installed, you should be able to setup the environment with the following command:
# $> conda env create -n question-generator python=3.12
# $> conda activate question-generator
# $> pip install magentic
# $> pip install pandas

# To use OpenAI, then you will need to set an environment variable in your shell called OPENAI_API_KEY with your API key.
# You can do this by running the following command in your terminal:
# $> export OPENAI_API_KEY=your-api-key
# $> python questions.py

# Then to run the example below if it were in a file called questions.py, you would run:
# $> python questions.py

# Standard Library Imports
from enum import Enum
from random import sample, shuffle, randint
from typing import List, Optional

# Third Party Imports (installed via pip in documentation above)
from magentic import prompt
from pandas import DataFrame
from pydantic import BaseModel


# Create an Enum with your topics of interest
class FocusArea(str, Enum):
    MATH = "math"
    SCIENCE = "science"
    HISTORY = "history"
    LITERATURE = "literature"
    GEOGRAPHY = "geography"
    COMPUTER_SCIENCE = "computer_science"
    ART = "art"
    MUSIC = "music"

    def __str__(self):
        return self.value

    def __repr__(self):
        return self.value


# Create a function to generate a list of focus areas at a desired distribution
def generate_focus_areas(n: int):
    all_areas = list(FocusArea)
    num_areas = len(all_areas)
    base_count = n // num_areas
    remainder = n % num_areas
    result = [area for area in all_areas for _ in range(base_count)]
    if remainder > 0:
        extra_areas = sample(all_areas, remainder)
        result.extend(extra_areas)
    shuffle(result)
    return result


# Create an object to represent the question 
# Properties here will be the columns in CSV output
class FindTheIncorrectStatementQuestion(BaseModel):
    question: str
    correct_answers: List[str]
    incorrect_answer: str
    difficulty: int
    topic: FocusArea
    explanation: Optional[str]


# Use Magentic `prompt` decorator to interact with LLM
@prompt('''
Your Task:
Create a single, university-level question based on the specified topic at a difficulty level based on the difficulty index value provided. The question should present four statements, three of which are correct and one of which is incorrect. The student must identify the incorrect statement.

Requirements:

Topic: {focus_area}
Difficulty: {difficulty} out of 10
Provide 3 correct statements about the topic {focus_area}, each one a complete and accurate sentence.
Provide 1 incorrect statement, also a complete sentence, indistinguishable in sentence structure from companion statements, that seems plausible but is actually false or misleading.
After listing all four statements, prompt the student to identify which one is incorrect.

Include an explanation section after the statements that clearly explains why the incorrect statement is wrong and how it could be misleading.

Explanation:
[Explain why the incorrect statement is wrong and misleading, providing any necessary clarifications or corrections.]''')
def generate_question(focus_area: FocusArea, difficulty: int) -> FindTheIncorrectStatementQuestion:
    ...


N = 15  # How many questions to generate

focus_areas = generate_focus_areas(N)
difficulties = [randint(1, 10) for _ in range(N)]
questions = map(generate_question, focus_areas, difficulties)
questions_df = DataFrame([q.model_dump() for q in questions])
questions_df.to_csv("find_the_incorrect_statement_questions.csv", index=False)