nevergrad/optimization/recastlib.py [214:250]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        pymoooptimize.minimize(problem, algorithm, seed=weakself._initial_seed)
        return None

    def _internal_ask_candidate(self) -> p.Parameter:
        """
        Special version to make sure that num_objectives has been set before
        the proper _internal_ask_candidate, in our parent class, is called.
        """
        if self.num_objectives == 0:
            # dummy ask i.e. not activating pymoo until num_objectives is set
            warnings.warn(
                "with this optimizer, it is more efficient to set num_objectives before the optimization begins",
                errors.NevergradRuntimeWarning,
            )
            # We need to get a datapoint that is a random point in parameter space,
            # and waste an evaluation on it.
            return self.parametrization.spawn_child()
        return super()._internal_ask_candidate()

    def _internal_tell_candidate(self, candidate: p.Parameter, loss: float) -> None:
        """
        Special version to make sure that we the extra initial evaluation which
        we may have done in order to get num_objectives, is discarded.
        Note that this discarding means that the extra point will not make it into
        replay_archive_tell. Correspondingly, because num_objectives will make it into
        the pickle, __setstate__ will never need a dummy ask.
        """
        if self._messaging_thread is None:
            return  # dummy tell i.e. not activating pymoo until num_objectives is set
        super()._internal_tell_candidate(candidate, loss)

    def _post_loss(self, candidate: p.Parameter, loss: float) -> tp.Loss:
        # pylint: disable=unused-argument
        """
        Multi-Objective override for this function.
        """
        return candidate.losses
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



nevergrad/optimization/recastlib.py [342:371]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        pymoooptimize.minimize(problem, algorithm, seed=weakself._initial_seed)
        return None

    def _internal_ask_candidate(self) -> p.Parameter:
        """Reads messages from the thread in which the underlying optimization function is running
        New messages are sent as "ask".
        """
        # get a datapoint that is a random point in parameter space
        if self.num_objectives == 0:  # dummy ask i.e. not activating pymoo until num_objectives is set
            warnings.warn(
                "with this optimizer, it is more efficient to set num_objectives before the optimization begins",
                errors.NevergradRuntimeWarning,
            )
            return self.parametrization.spawn_child()
        return super()._internal_ask_candidate()

    def _internal_tell_candidate(self, candidate: p.Parameter, loss: float) -> None:
        """Returns value for a point which was "asked"
        (none asked point cannot be "tell")
        """
        if self._messaging_thread is None:
            return  # dummy tell i.e. not activating pymoo until num_objectives is set
        super()._internal_tell_candidate(candidate, loss)

    def _post_loss(self, candidate: p.Parameter, loss: float) -> tp.Loss:
        # pylint: disable=unused-argument
        """
        Multi-Objective override for this function.
        """
        return candidate.losses
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -