Is the calculation of the AIC not incorrect?
The definition of AIC is "-2 * (LogLik - number of params)". Here, I believe the LogLik in semopy is not converted into the log likelihood of the Wishart distribution unless it is multiplied by the sample size n. In fact, while model.mx_cov is the sample covariance matrix, the statistic that follows the Wishart distribution is the one that has been multiplied by the sample size n. I am sorry if I am wrong.
Edited by 内場崇之