Errors

Errors from BEAM’s computational core are raised as subclasses of BEAM.BeamError and of the built-in exception they correspond to, so both except BEAM.BeamError and the matching built-in (except ValueError, except MemoryError, …) catch them. A few argument checks made in Python, such as float() of a BoundedValue, raise the plain built-in (ValueError, TypeError). Catching the built-in always works; catch the BEAM classes when you want to tell refusals apart.

import BEAM

try:
    model.entropy()
except BEAM.BudgetExceeded as err:     # one specific refusal
    print(err)                         # the message says what was needed

Exception	Means	Also a
BeamError	Base of BEAM’s own refusals.	Exception
BeamValueError	An argument BEAM cannot use.	ValueError
InvalidSequence	A sequence BEAM cannot read: bad characters, empty, too long (for example in explain).	ValueError
BeamTypeError	An argument of the wrong type.	TypeError
NotFinalized	The model cannot be used this way, for example save() after compact().	RuntimeError
NoLivePaths	Nothing satisfies the condition asked for, so the conditional does not exist.	RuntimeError
ModelMismatch	Two models cannot be combined: different alphabets, orders or shapes.	RuntimeError
BudgetExceeded	Refused because it would exceed the memory budget (4 GiB unless you set one).	MemoryError
OutOfMemory	The machine could not supply the memory.	MemoryError
ModelFileError	A model file could not be read or written.	OSError
LikelihoodPieceError	One piece of a repertoire likelihood could not be computed; the total is unaffected. SaddlepointDidNotConverge is one kind.	RuntimeError
InternalError	A bug in BEAM; please report it.	RuntimeError

Queries (BeamQuery) add QueryError, ImpossibleCondition, UnsupportedQuery and QueryTooLarge; mixtures (Mixture) add MixtureError, IncompatibleMembers, UnsupportedOrder and MixtureTooLarge. ImpossibleCondition is also a NoLivePaths; UnsupportedQuery and UnsupportedOrder are BeamValueErrors; QueryTooLarge and MixtureTooLarge are BudgetExceeded; IncompatibleMembers is a ModelMismatch. All are importable from BEAM.

BudgetExceeded is not OutOfMemory

BudgetExceeded means BEAM declined because of the memory budget (4 GiB by default, set with BEAM.set_memory_budget); raise the budget, ask at a lower order, or ask a smaller question. OutOfMemory means the machine itself ran out, which a larger budget will not fix.

Zero is an answer, not an error

log_pseq never raises: a sequence it cannot read (empty, longer than max_length, or with a character outside the alphabet) scores -inf, a probability of zero. Every readable sequence, including one never seen, has a non-zero probability, and every length from 1 to max_length has some probability. What raises is a conditional on an impossible event, which does not exist.