def testSampleHyperspherePositiveQuadrant(self):
for d in range(1, 10):
self.assertTrue(
sample_hypersphere_positive_quadrant(d).norm().isclose(
torch.tensor([1.0], dtype=torch.double)),
"sampled hypersphere point's norm is not 1.0",
)
def _extract_random_scalarization_settings(
objective_weights: Tensor,
outcome_constraints: Optional[Tuple[Tensor, Tensor]] = None,
**kwargs: Any,
) -> Tensor:
"""Generate a random weighting based on scalarization settings."""
use_random_scalarization = kwargs.get("random_scalarization", False)
random_weights = None
if use_random_scalarization:
# Pareto Optimization incompatible with outcome constraints.
if outcome_constraints is not None:
raise ValueError(
"Random scalarization for pareto frontier exploration "
"is incompatible with outcome constraints. Remove one.")
# Set distribution and sample weights.
distribution = kwargs.get("random_scalarization_distribution", SIMPLEX)
if distribution == SIMPLEX:
random_weights = sample_simplex(len(objective_weights))
elif distribution == HYPERSPHERE:
random_weights = sample_hypersphere_positive_quadrant(
len(objective_weights))
if random_weights is not None:
objective_weights = torch.mul(objective_weights, random_weights)
return objective_weights