In this repository we will explore the capabilities of LLMs and ChatGPT for lithology prediction based on some well log measurements. First we will explore the data and use some vanilla machine learning approaches (logistic regression, SVM, random forest, XGBoost) and some new methods such as KANs. Then we will explore how LLMs compare with these baselines. In particular we will explore the following:
Conventional machine learning methods (SVM, XGBoost, Logistic Regression, KAN, Random Forest)
Using text embeddings from ChatGPT as inputs to conventional classifiers (hybrid methods)
RAG-based ChatGPT for lithology classification (pure language-based methods)
Using LLMs for tabular data is a very niche and promising area, and this article will explore some initial results to lithology prediction. Below are some references related to applications of LLMs in tabular data