Chimera Objective #455
Chimera Objective #455
Conversation
|
|
|
@AVHopp @AdrianSosic PR in draft mode, NOT for review |
| @@ -127,6 +128,12 @@ def to_botorch( | |||
| additional_params["best_f"] = ( | |||
| bo_surrogate.posterior(train_x).mean.max().item() | |||
| ) | |||
| case ChimeraObjective(): | |||
There was a problem hiding this comment.
move to above block utilizing minimization as discussed
|
|
||
| @define(frozen=True, slots=False) | ||
| class ChimeraObjective(Objective): | ||
| """An objective scalarizing multiple targets using desirability values.""" |
There was a problem hiding this comment.
wrong docstring mentioning desirability
incude a link to the publication, see other code parts (eg edbo) for how to properly include links
| ) | ||
| "The targets considered by the objective." | ||
|
|
||
| targets_threshold_values: tuple[float, ...] = field( |
There was a problem hiding this comment.
targets_ in fornt is not necessary for either of the two threshold attributes
| """The softness parameter regulating the Heaviside function.""" | ||
|
|
||
| @targets_threshold_values.default | ||
| def _default_targets_threshold_values(self) -> tuple[float, ...]: |
There was a problem hiding this comment.
i think there are no rasonable defaults so these should not have defaults
apart from that the defaults you specified do not satisfy the requirements ge(0.0)
| return default_values | ||
|
|
||
| @softness.default # TODO: do we need to add warning here? | ||
| def _default_softness(self) -> float: |
There was a problem hiding this comment.
this can simply be done by writing field(..., default=1e-3, ...) above where the attribute is specified
no warning needed
| if not all(target._is_transform_normalized for target in targets): | ||
| raise ValueError( | ||
| "All targets must have normalized computational representations to " | ||
| "enable the computation of desirability values. This requires having " |
There was a problem hiding this comment.
update message, still mentions desirability
| arg = -value / softness | ||
| return np.exp(-np.logaddexp(0, arg)) | ||
|
|
||
| def _hard_heaviside(self, value: float) -> float: |
There was a problem hiding this comment.
is the hard function really needed? I think the soft heavyside should just recover the hard one for an extreme value of softness. If so that can simply be sued wherever a hard heaviside is needed
|
|
||
| return self._soft_heaviside(value, softness) | ||
|
|
||
| def _invert_binary(self, a: float) -> float: |
There was a problem hiding this comment.
this seems so small and only used once, is this really needed?
| for target, threshold_value, threshold_type in zip( | ||
| self.targets, transformed_threshold_values, self.targets_threshold_types | ||
| ): | ||
| if threshold_type == ThresholdType.FRACTION: |
There was a problem hiding this comment.
all of these == checks should instead be done with is. The possible enum values are singletons and they should be checked via is jsut as for checks of soemthing is None
| threshold_value, interpolation="linear" | ||
| ) | ||
| elif threshold_type == ThresholdType.ABSOLUTE: | ||
| _threshold = threshold_value |
There was a problem hiding this comment.
why is fraction and absolute treated in the exact same way?
This PR introduces Chimera, a general-purpose achievement scalarizing function for multi-target optimization that allows users to establish a hierarchy of targets with relative or absolute thresholds for concurrent optimization.
This implementation includes a new objective class,
ChimeraObjective, which follows a similar approach to theDesirabilityObjective. It scalarizes multiple targets into a single score, termedChimera Merits, which is to be minimized.For further details, please refer to the following publication:
F. Häse, L.M. Roch, and A. Aspuru-Guzik. Chimera: enabling hierarchy-based multi-objective optimization for self-driving laboratories. Chemical Science 2018, 9(39), 7642-7655.
WIP:
ChimeraObjectiveimplementation.DesirabilityObjective.ChimeraObjective.