Hello!
I'm wondering how did authors select the criterion of fixing symbolic function of activation node as correlation coefficient.
In Symbolic_LANLayer.fix_symbolic() method, I found that r2 is calculated by utils.fit_params() function, and fit_params() function derives r2 value between activation node and symbolic function using correlation between them.
I think that it is more appropriate to use r2 as coefficient of determination(R^2) value as r2 variable in code, so I'm wondering why authors select correlation as r2 between two functions. Is there any reason?
Hello! I'm wondering how did authors select the criterion of fixing symbolic function of activation node as correlation coefficient. In Symbolic_LANLayer.fix_symbolic() method, I found that r2 is calculated by utils.fit_params() function, and fit_params() function derives r2 value between activation node and symbolic function using correlation between them. I think that it is more appropriate to use r2 as coefficient of determination(R^2) value as r2 variable in code, so I'm wondering why authors select correlation as r2 between two functions. Is there any reason?
Thank you for any opinion for the question!