Remove `joint_logprob` function from `tests.logprob.util` submodule

[ricardoV94]

Description

This function is not very helpful and only used in tests. We should be able to remove it, and use either logp or factorized_joint_logprob in the tests.

pymc/tests/logprob/utils.py

Lines 50 to 74 in 83cd926

	def joint_logprob(*args, sum: bool = True, **kwargs) -> Optional[TensorVariable]:
	"""Create a graph representing the joint log-probability/measure of a graph.
	This function calls `factorized_joint_logprob` and returns the combined
	log-probability factors as a single graph.
	Parameters
	----------
	sum: bool
	If ``True`` each factor is collapsed to a scalar via ``sum`` before
	being joined with the remaining factors. This may be necessary to
	avoid incorrect broadcasting among independent factors.
	"""
	logprob = factorized_joint_logprob(*args, **kwargs)
	if not logprob:
	return None
	if len(logprob) == 1:
	logprob = tuple(logprob.values())[0]
	if sum:
	return pt.sum(logprob)
	return logprob
	if sum:
	return pt.sum([pt.sum(factor) for factor in logprob.values()])
	return pt.add(*logprob.values())

[shreyas3156]

Hi @ricardoV94, I'd like to take this issue up and get a better idea of logprob-related modules in the process.

There is a snippet inside the logp() method in model.py where joint_logp() is used.

pymc/pymc/model.py

Lines 758 to 765 in 2fcce43

	if rvs:
	rv_logps = joint_logp(
	rvs=rvs,
	rvs_to_values=self.rvs_to_values,
	rvs_to_transforms=self.rvs_to_transforms,
	jacobian=jacobian,
	)
	assert isinstance(rv_logps, list)

Should the logic from joint_logp() simply be added to the above section itself, to handle this single usage?

[ricardoV94]

@shreyas3156 Apologies, I referenced the wrong function. The one that should be removed is this one from the tests utils:

pymc/tests/logprob/utils.py

Lines 50 to 74 in 83cd926

	def joint_logprob(*args, sum: bool = True, **kwargs) -> Optional[TensorVariable]:
	"""Create a graph representing the joint log-probability/measure of a graph.
	This function calls `factorized_joint_logprob` and returns the combined
	log-probability factors as a single graph.
	Parameters
	----------
	sum: bool
	If ``True`` each factor is collapsed to a scalar via ``sum`` before
	being joined with the remaining factors. This may be necessary to
	avoid incorrect broadcasting among independent factors.
	"""
	logprob = factorized_joint_logprob(*args, **kwargs)
	if not logprob:
	return None
	if len(logprob) == 1:
	logprob = tuple(logprob.values())[0]
	if sum:
	return pt.sum(logprob)
	return logprob
	if sum:
	return pt.sum([pt.sum(factor) for factor in logprob.values()])
	return pt.add(*logprob.values())