AssertionError: Invalid progress was set by the TerminationCriterion

[dinohsu1019]

Hi Julian,

I see this error after 2 iterations and 90 evaluations for a 4-objective, 32-solution-varialbe (integers) problem, and I didn't see it last time with the same termination and minimize loop code. Looking at calss DefaultTermination:

class DefaultTermination(Termination):

    def __init__(self, x, cv, f, n_max_gen=1000, n_max_evals=100000) -> None:
        super().__init__()
        self.x = x
        self.cv = cv
        self.f = f

        self.max_gen = MaximumGenerationTermination(n_max_gen)
        self.max_evals = MaximumFunctionCallTermination(n_max_evals)

        self.criteria = [self.x, self.cv, self.f, self.max_gen, self.max_evals]

    def _update(self, algorithm):
        p = [criterion.update(algorithm) for criterion in self.criteria]
        return max(p)

It seems something to do the criterion.update method, but I don't know why self.f, etc. has an update method.
Below are my codes, please shed some light, thanks.

# NSGA-II algorithm setup
algorithm = NSGA2(
    pop_size = POP_SIZE,
    n_offsprings = N_OFFSPRINGS,
    sampling=IntegerRandomSampling(),
    crossover=SBX(prob=SBX_PROB, eta=SBX_ETA, vtype=float, repair=RoundingRepair()),
    mutation=PM(prob=PM_PROB, eta=PM_ETA, vtype=float, repair=RoundingRepair()),
    eliminate_duplicates = ELIMINATE_DUPLICATES
)

termination = DefaultMultiObjectiveTermination(
    ftol = FTOL,
    period = PERIOD,
    n_max_gen = N_MAX_GEN,
    n_max_evals = N_MAX_EVALS
)

solver = algorithm.setup(
    problem,
    termination = termination,
    seed = 1,
    verbose = True,
    save_history = True
)

hv = HV(ref_point=REF_POINT)

while solver.has_next():
    
    solver.next()

    n_gen = solver.n_gen-1
    n_eval = solver.evaluator.n_eval
    nds_F = solver.opt.get("F")
    nds_X = solver.opt.get("X")
    # opt_F, opt_X, opt_index = get_my_opt(nds_F, nds_X)
    # opt_F, opt_X, opt_index = get_ASF_opt(nds_F, nds_X, WEIGHTS)
    opt_F, opt_X, opt_index = get_TNeagle_opt(nds_F, nds_X, IDEAL_MDD)

    hypervol = hv.do(nds_F)
    
    with out_gen_perf:
        print(f"{n_gen=} {n_eval=} {opt_F=} {opt_X=} {opt_index=} {hypervol=}")

solver.Finalize()
res = solver.result()
n_gen = solver.n_gen-1
n_eval = solver.evaluator.n_eval
exec_time = res.exec_time

Here are the error messages:

---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
Cell In[1], line 577
    573 hv = HV(ref_point=REF_POINT)
    575 while solver.has_next():
--> 577     solver.next()
    579     n_gen = solver.n_gen-1
    580     n_eval = solver.evaluator.n_eval

File ~\miniconda3\envs\finlab\lib\site-packages\pymoo\core\algorithm.py:159, in Algorithm.next(self)
    157 if infills is not None:
    158     self.evaluator.eval(self.problem, infills, algorithm=self)
--> 159     self.advance(infills=infills)
    161 # if the algorithm does not follow the infill-advance scheme just call advance
    162 else:
    163     self.advance()

File ~\miniconda3\envs\finlab\lib\site-packages\pymoo\core\algorithm.py:229, in Algorithm.advance(self, infills, **kwargs)
    227     # always advance to the next iteration - except if the algorithm returns False
    228     if val is None or val:
--> 229         self._post_advance()
    231 # if the algorithm has terminated, then do the finalization steps and return the result
    232 if self.termination.has_terminated():

File ~\miniconda3\envs\finlab\lib\site-packages\pymoo\core\algorithm.py:309, in Algorithm._post_advance(self)
    306 self.termination.update(self)
    308 # display the output if defined by the algorithm
--> 309 self.display(self)
    311 # if a callback function is provided it is called after each iteration
    312 self.callback(self)

File ~\miniconda3\envs\finlab\lib\site-packages\pymoo\core\callback.py:27, in Callback.__call__(self, algorithm)
     24     self.is_initialized = True
     26 self.notify(algorithm)
---> 27 self.update(algorithm)

File ~\miniconda3\envs\finlab\lib\site-packages\pymoo\util\display\display.py:19, in Display.update(self, algorithm, **kwargs)
     17 text = ""
     18 header = not output.is_initialized
---> 19 output(algorithm)
     21 if header:
     22     text += output.header(border=True) + '\n'

File ~\miniconda3\envs\finlab\lib\site-packages\pymoo\core\callback.py:27, in Callback.__call__(self, algorithm)
     24     self.is_initialized = True
     26 self.notify(algorithm)
---> 27 self.update(algorithm)

File ~\miniconda3\envs\finlab\lib\site-packages\pymoo\util\display\multi.py:79, in MultiObjectiveOutput.update(self, algorithm)
     76 if self.indicator_no_pf is not None:
     78     ind = self.indicator_no_pf
---> 79     ind.update(algorithm)
     81     valid = ind.delta_ideal is not None
     83     if valid:

File ~\miniconda3\envs\finlab\lib\site-packages\pymoo\core\termination.py:29, in Termination.update(self, algorithm)
     27 else:
     28     progress = self._update(algorithm)
---> 29     assert progress >= 0.0, "Invalid progress was set by the TerminationCriterion."
     31 self.perc = progress
     32 return self.perc

AssertionError: Invalid progress was set by the TerminationCriterion.

[dinohsu1019]

A quick update, this error seems to be caused by objective values as np.nan, I have modified the _evaluate method to avoid it, will see if this problem still occurs, thanks.

[blankjul]

Great you could resolve. Please next time also post code to reproduce your error (with a test problem and not your whole application if possible).