feature/libe_gen_wrapper + Feature/asktell aposmm #209
Conversation
for more information, see https://pre-commit.ci
… feature/asktell_aposmm
…etup to start the thread
for more information, see https://pre-commit.ci
…overwriting x with new f values? probably a bug on my end
…g results_array to be empty
…imas into feature/asktell_aposmm
for more information, see https://pre-commit.ci
…imas into feature/asktell_aposmm
for more information, see https://pre-commit.ci
…lot in points to variably-sized results-array (based on either initial sample, or subsequent points)
for more information, see https://pre-commit.ci
optimas/generators/libE_wrapper.py
Outdated
| self.libe_gen = self.libe_gen_class( | ||
| gen_specs, H, persis_info, libE_info | ||
| ) | ||
| elif self.libe_gen_instance is not None: |
There was a problem hiding this comment.
I don't see how libe_gen_instance is different from libe_gen. Seems redundant.
There was a problem hiding this comment.
It may be. I can do a bunch of refactoring now that we have this working
| dist_to_bound_multiple=0.5, | ||
| max_active_runs=4, # refers to APOSMM's simul local optimization runs | ||
| lb=np.array([-3, -2]), # potentially matches the VaryingParameters | ||
| ub=np.array([3, 2]), |
There was a problem hiding this comment.
lb/ub duplicated from VaryingParameter. Not sure best approach, but for now we could define these first and use in both places.
There was a problem hiding this comment.
Yup! Perhaps our generator_standard work may help this avenue.
There was a problem hiding this comment.
Insert same source-of-truth values into VaryingParameters and ub and lb
| gen = libEWrapper( | ||
| varying_parameters=[var_1, var_2], | ||
| objectives=[obj], | ||
| libe_gen_instance=aposmm, |
There was a problem hiding this comment.
How different to libe_gen. This can be set up so if a class is provided it initializes in place.
e.g.,
if inspect.isclass(libe_gen):
initialize...
There was a problem hiding this comment.
That's true, for rough-development I was being explicit on what type of object we passed, but we can afford to be smarter now. Pass in something initialized, keep as-is, if its not initialized, initialize it.
I'll take a second look, but I don't think I came across a pure-optimas example that did object-initialization within the library instead of exposed to the user.
… cleanup, better handling of non-existance of final_tell
for more information, see https://pre-commit.ci
…imas into feature/asktell_aposmm
for more information, see https://pre-commit.ci
…hod. Remove now-unneeded gen reference from wrapper, since serialization issues for live-gen. *But*, the reference stays alive in libE
…imas into feature/asktell_aposmm
for more information, see https://pre-commit.ci
…l TrialParameters, plus all aposmm-specific array quirks.
for more information, see https://pre-commit.ci
| exp.run(100) | ||
| exp.run(200) | ||
| exp.finalize() | ||
| assert len(gen.libe_gen.all_local_minima) |
There was a problem hiding this comment.
Do we want to check the minima have the expected values.
…itional docs for Exploration.finalize()
for more information, see https://pre-commit.ci
| lb=np.array([var_1.lower_bound, var_2.lower_bound]), | ||
| ub=np.array([var_1.upper_bound, var_2.upper_bound]), |
There was a problem hiding this comment.
Could we replace this with:
variables = {'x0':[-3.0, 3.0], 'x1':[-2.0, 2.0]},
objectives = {'f': 'MINIMIZE'}
as discussed here: campa-consortium/generator_standard#16
Depends on upstream libEnsemble branch: experimental/jlnav_plus_shuds_asktell