PymcSampler and fixed parameters

[dweindl]

Something's fishy with the handling of fixed parameters in PymcSampler.

PymcSampler.initialize stores a reduced x0:

pyPESTO/pypesto/sample/pymc.py

Line 184 in fd976b7

x0 = problem.get_reduced_vector(x0)

Whereas PymcSampler.sample expects a full x0:

pyPESTO/pypesto/sample/pymc.py

Line 211 in fd976b7

for x_name, val in zip(problem.x_names, self.x0)

Not sure about the impact, but seems like this might have quite unintended consequences.

Would be great if somebody could look into that, fix it, and document the expected length of PymcSampler.x0

To reproduce: add strict=True to

pyPESTO/pypesto/sample/pymc.py

Line 211 in fd976b7

for x_name, val in zip(problem.x_names, self.x0)

and run test/sample/test_sample.py::test_pipeline[Pymc-rosenbrock]

[dweindl]

Same problem in PymcVariational.fit:

pyPESTO/pypesto/variational/pymc.py

Line 85 in fd976b7

for x_name, val in zip(problem.x_names, self.x0)

[PaulJonasJost]

great catch 👍🏼 Sometimes puzzled by some default behaviour, would have honestly expected zip() to throw an error with unequal lengths 🙈

[PaulJonasJost]

as this feels very much connected to #1682, should we switch to strict=True everywhere?

[PaulJonasJost]

as this feels very much connected to #1682, should we switch to strict=True everywhere?

already initiated in #1679