I saw the task2vec complexity using l1. But the l1 complexities might not be comparable across data sets. How would you suggest to adjust this metric s.t. its directly comparable accross data sets?
My suggestion is to divide perhaps by the std l1 complexity for that data set e.g.:
standardized_complexity = avg_complexity(task2vecs_list, l1) / unbiased_std_complexity(task2vecs_list, l1)
thoughts? This does assume normality and n>=30. Histograms showing normality might be useful.
related: https://stats.stackexchange.com/questions/604296/how-does-one-create-comparable-metrics-when-the-original-metrics-are-not-compara
I saw the task2vec complexity using l1. But the l1 complexities might not be comparable across data sets. How would you suggest to adjust this metric s.t. its directly comparable accross data sets?
My suggestion is to divide perhaps by the std l1 complexity for that data set e.g.:
thoughts? This does assume normality and n>=30. Histograms showing normality might be useful.
related: https://stats.stackexchange.com/questions/604296/how-does-one-create-comparable-metrics-when-the-original-metrics-are-not-compara