✨ [EMNLP2025] On the Role of Model Prior in Real-World Inductive Reasoning

🔬 Paper

🧩 Overview

This project explores how model priors affect real-world inductive reasoning tasks.
We provide experiments demonstrating how different model configurations and data setups impact reasoning performance under various conditions.

🚀 Usage

1. Clone the Repository

git clone https://github.com/joeliuz6/Model-prior-in-inductive-reasoning.git
cd Model-prior-in-inductive-reasoning

2. Create and Activate Environment

conda create -n modelprior python=3.10
conda activate modelprior

3. Set Your OpenAI API Key

Linux / macOS:

export OPENAI_API_KEY="your_api_key_here"

Windows (PowerShell):

setx OPENAI_API_KEY "your_api_key_here"

You need to also fill in the client(api_key="").

4. Run main.py

With demonstrations and label configurations:

python main.py --with_data your_choice --data_label your_choice

5. Arguments Explanation

Argument	Type	Description
--with_data	str	Include in-context demonstrations in generation.
--data_label	str	Specify label configuration for different tasks.
--seed	int	(Optional) Random seed for reproducibility. Default: 2
--model	str	(Optional) Model name, e.g. "gpt-4o-2024-05-13".

Example:

python main.py --with_data no_data --data_label correct

Dataset

Full datasets in the paper. Google Drive

✏️ Citation

@article{liu2024role,
  title={On the Role of Model Prior in Real-World Inductive Reasoning},
  author={Liu, Zhuo and Yu, Ding and He, Hangfeng},
  journal={arXiv preprint arXiv:2412.13645},
  year={2024}
}