MFCK and SMFK compatible with heteroscedastic noise

[mcastanoUQ]

Updates to the classes MFCK, SMFCK, SMFK
Small correction of a typo in MFK.
MFCK, SMFCK updates to handle het. noise.

[relf]

Hi Mauricio. It seems like SMFCK tests (test_smfck.py) runs for ever! Then github actions are just cancelled.

Did you run your tests locally?

pytest smt\applications\tests\tests\test_smfck.py

[mcastanoUQ]

I see the issue; it's strange, I compiled the script outside the test class, and it runs without problems. I'll try to change the test script to see if that solves the problem for the moment.

[relf]

Hi @mcastanoUQ, smt does not have nlopt as a dependency, could you use cobyla from scipy.

[mcastanoUQ]

@relf I changed it locally, but now I have found a problem in the class lhs.py line 367

File "D:\mcastano\Documents\Repositories\smt\smt\sampling_methods\lhs.py", line 367, in _ese
else lhs(dim, nt, criterion=None, seed=self.random_state)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: lhs() got an unexpected keyword argument 'seed'

[mcastanoUQ]

If I change the line 367 from this:

else lhs(dim, nt, criterion=None, seed=self.random_state)

to this:

else lhs(dim, nt, criterion=None) #(I think it's not a solution, just a comment)

I can run everything without any problem. But, I think the issue is related to the change of "seed" instead of "random_state"

Should I change my code form "random_state" to "seed" everywhere, or does it work the same?

[relf]

Please update the PR without using nlopt and we will see.

Regarding the error related to LHS, you may need to update the pyDOE3 package locally with pip install -U pyDOE3.
random_state option is deprecated (as np.RandomState is deprecated), it should still work though but please use seed now (it uses np.Generator behind the scene).

[mcastanoUQ]

Bonjour @relf,

Une erreur ici: FAILED smt/tests/test_all.py::Test::test_exp_MGP - AssertionError: np.float64(1.781220778809592) not less than or equal to 0.06

Ce n’est pas parmi les classes que je suis en train de modifier ; je ne sais pas si c’est dû à un rebase manquant.

[relf]

Not sure to understand this docstring. You intend to make SMFCK inherits from MFCK in a further PR? Change it TODO: This class should inherit from the class MFCK

[relf]

You relax the test tolerance to make it pass with scipy cobyla ? Add a comment here to explain that one has to install nlopt to use Cobyla-nlopt to get better performance. Maybe add keep the old line self.assert_error(t_error, 0.0, 2e-1, 2e-1) to tell it works with Cobyla from no=lopt.