Is there any way to extract the likelihood estimation of a trained model ? Are there any measure for model comparison (e.g. AIC) ? Thank you