The =fit_graph= is now lightyears faster than the previous version
(redundant copying was removed)
If the number of observations is small, the underlying probability
tables tends to be sparse. In these cases the usual penalty term of AIC
and BIC is often too restrictive. One can now set =sparse_qic= to
=TRUE=. This penality is computed according to a sparse criteria. The
criteria resembles the usual penalty as the number of observations
grow.
The =q= parameter (0 for AIC and 1 for BIC) can now be any positive
number. The higher, the more penalty and the fewer edges will be
added.