Experiences of using GPT-4 to create a machine learning model for EDA tool license need forecasting
Poikolainen, Tommi (2024)
2024
Tietojohtamisen DI-ohjelma - Master's Programme in Information and Knowledge Management
Johtamisen ja talouden tiedekunta - Faculty of Management and Business
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2024-06-24
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202405286380
https://urn.fi/URN:NBN:fi:tuni-202405286380
Tiivistelmä
This thesis presents a method to predict the need for electronic design automation (EDA) tool licenses using generative pre-trained transformer 4 (GPT-4). The goal is to generate machine learning (ML) code that can forecast how many EDA tool licenses will be required. This process is crucial for planning and resource allocation in system on a chip program development.
The research follows an approach utilizing design science research, focusing on refining the prompts given to GPT-4 to improve the ML code it generates. This involves creating scripts for preparing data, training ML models, and using these models, all tailored to predict EDA tool license usage accurately. The study is grounded in knowledge from multiple areas, including ML, GPT-4's capabilities, forecasting models, and how to evaluate their effectiveness.
The thesis aims to produce a set of guidelines and best practices for using GPT-4 to create ML code for EDA tool license forecasting. It seeks to show how GPT-4 can be a useful tool in developing advanced forecasting methods for evaluating future EDA tool needs.
ML code generation with GPT-4 can be deemed possible with a carefully planned process, however, the results also indicate that the ML models' predictions were not accurate enough for practical use. This is attributed to insufficient hyperparameter tuning, a lack of data, and the need for more detailed features within the data. These findings highlight the challenges of using GPT-4 for generating functional ML code in this context and suggest areas for further research, including improving data quality and model tuning.
The research follows an approach utilizing design science research, focusing on refining the prompts given to GPT-4 to improve the ML code it generates. This involves creating scripts for preparing data, training ML models, and using these models, all tailored to predict EDA tool license usage accurately. The study is grounded in knowledge from multiple areas, including ML, GPT-4's capabilities, forecasting models, and how to evaluate their effectiveness.
The thesis aims to produce a set of guidelines and best practices for using GPT-4 to create ML code for EDA tool license forecasting. It seeks to show how GPT-4 can be a useful tool in developing advanced forecasting methods for evaluating future EDA tool needs.
ML code generation with GPT-4 can be deemed possible with a carefully planned process, however, the results also indicate that the ML models' predictions were not accurate enough for practical use. This is attributed to insufficient hyperparameter tuning, a lack of data, and the need for more detailed features within the data. These findings highlight the challenges of using GPT-4 for generating functional ML code in this context and suggest areas for further research, including improving data quality and model tuning.