Data feature desgin

try to predict the hyperparameter of data scaling law with dataset features like zipf or sth else