A Data Centric Perspective On Evaluating Machine Learning Models For Tabular Data