In previous versions, the Smoothed Probability Estimation could be defined in the overall program options.
Now, instead, it can be set for each graph, i.e. for each xbl file, individually.
The defined virtual occurrences are spread uniformly over all the joint probability distribution.
Thanks to those virtual occurrences, it is then impossible to have any deterministic relations, even if parents-children combinations are not present in the dataset.
This option is particularly useful for supervised learning, as it generally increases the prediction quality of the model (better generalization).