Functional Validation of Supervised Learning Tasks for Reproducible Machine Learning

Ensuring reproducibility in Machine Learning (ML) research is crucial for scientific integrity and advancing the field. However, reproducing exact results for models is a significant challenge due to factors such as data availability, training model of interest, and resource discrepancies. To overcome this problem, we developed a framework for researchers to show that their methods generate the claimed results by running our FunctionalModel class. Our approach addresses challenges related to dataset distribution, model saving, metrics tampering, and suggesting an automatic process to validate the results. In this paper, we focus on the first stage of reproducibility, which is functional evaluation to validate or provide evidence whether the claimed outcomes are actually obtained or not while maintaining the data. Our approach enhances transparency, accountability, and trustworthiness in ML research.