from msmtools.estimation.dense.estimate_transition_matrix_reversible
from msmtools.estimation.sparse.mle_trev.mle_trev
have different defaults for the maxerr parameter. This results in the fact that for some count matrices one implementation will converge, the other one won't converge with the default setting, and what's worse you will get different results.
What about defining the convergence parameters explicitly on the API level (msmtools.estimation.transition_matrix)? In nonreversible estimation we just don't use them, in the reversible case we pass them on. Avoids inconsistencies and also makes the method more user-friendly, because autocomplete will show important parameters.
The two methods
have different defaults for the
maxerr
parameter. This results in the fact that for some count matrices one implementation will converge, the other one won't converge with the default setting, and what's worse you will get different results.What about defining the convergence parameters explicitly on the API level (msmtools.estimation.transition_matrix)? In nonreversible estimation we just don't use them, in the reversible case we pass them on. Avoids inconsistencies and also makes the method more user-friendly, because autocomplete will show important parameters.