def test_apply_constraints_multi_output(self):
# nonnegative objective, one constraint
tkwargs = {"device": self.device}
for dtype in (torch.float, torch.double):
tkwargs["dtype"] = dtype
samples = torch.rand(3, 2, **tkwargs)
obj = samples.clone()
obj = apply_constraints(obj=obj,
constraints=[zeros_f],
samples=samples,
infeasible_cost=0.0)
self.assertTrue(torch.equal(obj, samples * 0.5))
# nonnegative objective, two constraint
obj = samples.clone()
obj = apply_constraints(
obj=obj,
constraints=[zeros_f, zeros_f],
samples=samples,
infeasible_cost=0.0,
)
self.assertTrue(torch.equal(obj, samples * 0.5 * 0.5))
# negative objective, one constraint, infeasible_cost
obj = samples.clone().clamp_min(-1.0)
obj = apply_constraints(obj=obj,
constraints=[zeros_f],
samples=samples,
infeasible_cost=2.0)
self.assertTrue(
torch.allclose(obj,
samples.clamp_min(-1.0) * 0.5 - 1.0))
# nonnegative objective, one constraint, eta = 0
obj = samples
with self.assertRaises(ValueError):
apply_constraints(
obj=obj,
constraints=[zeros_f],
samples=samples,
infeasible_cost=0.0,
eta=0.0,
)
def test_apply_constraints(self):
# nonnegative objective, one constraint
samples = torch.randn(1)
obj = ones_f(samples)
obj = apply_constraints(obj=obj,
constraints=[zeros_f],
samples=samples,
infeasible_cost=0.0)
self.assertTrue(torch.equal(obj, ones_f(samples) * 0.5))
# nonnegative objective, two constraint
samples = torch.randn(1)
obj = ones_f(samples)
obj = apply_constraints(
obj=obj,
constraints=[zeros_f, zeros_f],
samples=samples,
infeasible_cost=0.0,
)
self.assertTrue(torch.equal(obj, ones_f(samples) * 0.5 * 0.5))
# negative objective, one constraint, infeasible_cost
samples = torch.randn(1)
obj = minus_one_f(samples)
obj = apply_constraints(obj=obj,
constraints=[zeros_f],
samples=samples,
infeasible_cost=2.0)
self.assertTrue(torch.equal(obj, ones_f(samples) * 0.5 - 2.0))
# nonnegative objective, one constraint, eta = 0
samples = torch.randn(1)
obj = ones_f(samples)
with self.assertRaises(ValueError):
apply_constraints(
obj=obj,
constraints=[zeros_f],
samples=samples,
infeasible_cost=0.0,
eta=0.0,
)
def forward(self, samples: Tensor) -> Tensor:
r"""Evaluate the feasibility-weighted objective on the samples.
Args:
samples: A `sample_shape x batch_shape x q x m`-dim Tensors of
samples from a model posterior.
Returns:
A `sample_shape x batch_shape x q`-dim Tensor of objective values
weighted by feasibility (assuming maximization).
"""
obj = super().forward(samples=samples)
return apply_constraints(
obj=obj,
constraints=self.constraints,
samples=samples,
infeasible_cost=self.infeasible_cost,
eta=self.eta,
)
def forward(self, samples: Tensor) -> Tensor:
r"""Evaluate the feasibility-weighted objective on the samples.
Args:
samples: A `sample_shape x batch_shape x q x o`-dim Tensors of
samples from a model posterior.
Returns:
A `sample_shape x batch_shape x q`-dim Tensor of objective values
weighted by feasibility (assuming maximization).
"""
obj = super().forward(samples=samples)
return apply_constraints(
obj=obj,
constraints=self.constraints,
samples=samples,
infeasible_cost=self.infeasible_cost,
eta=self.eta,
)
def forward(self, samples: Tensor, X: Optional[Tensor] = None) -> Tensor:
r"""Evaluate the feasibility-weighted objective on the samples.
Args:
samples: A `sample_shape x batch_shape x q x m`-dim Tensors of
samples from a model posterior.
X: A `batch_shape x q x d`-dim tensor of inputs. Relevant only if
the objective depends on the inputs explicitly.
Returns:
A `sample_shape x batch_shape x q`-dim Tensor of objective values
weighted by feasibility (assuming maximization).
"""
obj = super().forward(samples=samples)
return apply_constraints(
obj=obj,
constraints=self.constraints,
samples=samples,
infeasible_cost=self.infeasible_cost,
eta=self.eta,
)
def test_constrained_mc_objective(self):
for dtype in (torch.float, torch.double):
# one feasible constraint
obj = ConstrainedMCObjective(
objective=generic_obj, constraints=[feasible_con]
)
samples = torch.randn(1, device=self.device, dtype=dtype)
constrained_obj = generic_obj(samples)
constrained_obj = apply_constraints(
obj=constrained_obj,
constraints=[feasible_con],
samples=samples,
infeasible_cost=0.0,
)
self.assertTrue(torch.equal(obj(samples), constrained_obj))
# one infeasible constraint
obj = ConstrainedMCObjective(
objective=generic_obj, constraints=[infeasible_con]
)
samples = torch.randn(2, device=self.device, dtype=dtype)
constrained_obj = generic_obj(samples)
constrained_obj = apply_constraints(
obj=constrained_obj,
constraints=[infeasible_con],
samples=samples,
infeasible_cost=0.0,
)
self.assertTrue(torch.equal(obj(samples), constrained_obj))
# one feasible, one infeasible
obj = ConstrainedMCObjective(
objective=generic_obj, constraints=[feasible_con, infeasible_con]
)
samples = torch.randn(2, 1, device=self.device, dtype=dtype)
constrained_obj = generic_obj(samples)
constrained_obj = apply_constraints(
obj=constrained_obj,
constraints=[feasible_con, infeasible_con],
samples=samples,
infeasible_cost=0.0,
)
self.assertTrue(torch.equal(obj(samples), constrained_obj))
# one feasible, one infeasible, infeasible_cost
obj = ConstrainedMCObjective(
objective=generic_obj,
constraints=[feasible_con, infeasible_con],
infeasible_cost=5.0,
)
samples = torch.randn(3, 2, device=self.device, dtype=dtype)
constrained_obj = generic_obj(samples)
constrained_obj = apply_constraints(
obj=constrained_obj,
constraints=[feasible_con, infeasible_con],
samples=samples,
infeasible_cost=5.0,
)
self.assertTrue(torch.equal(obj(samples), constrained_obj))
# one feasible, one infeasible, infeasible_cost, higher dimension
obj = ConstrainedMCObjective(
objective=generic_obj,
constraints=[feasible_con, infeasible_con],
infeasible_cost=5.0,
)
samples = torch.randn(4, 3, 2, device=self.device, dtype=dtype)
constrained_obj = generic_obj(samples)
constrained_obj = apply_constraints(
obj=constrained_obj,
constraints=[feasible_con, infeasible_con],
samples=samples,
infeasible_cost=5.0,
)
self.assertTrue(torch.equal(obj(samples), constrained_obj))
def test_constrained_mc_objective(self, cuda=False):
device = torch.device("cuda") if cuda else torch.device("cpu")
for dtype in (torch.float, torch.double):
# one feasible constraint
obj = ConstrainedMCObjective(
objective=generic_obj, constraints=[feasible_con]
)
samples = torch.randn(1, device=device, dtype=dtype)
constrained_obj = generic_obj(samples)
constrained_obj = apply_constraints(
obj=constrained_obj,
constraints=[feasible_con],
samples=samples,
infeasible_cost=0.0,
)
self.assertTrue(torch.equal(obj(samples), constrained_obj))
# one infeasible constraint
obj = ConstrainedMCObjective(
objective=generic_obj, constraints=[infeasible_con]
)
samples = torch.randn(2, device=device, dtype=dtype)
constrained_obj = generic_obj(samples)
constrained_obj = apply_constraints(
obj=constrained_obj,
constraints=[infeasible_con],
samples=samples,
infeasible_cost=0.0,
)
self.assertTrue(torch.equal(obj(samples), constrained_obj))
# one feasible, one infeasible
obj = ConstrainedMCObjective(
objective=generic_obj, constraints=[feasible_con, infeasible_con]
)
samples = torch.randn(2, 1, device=device, dtype=dtype)
constrained_obj = generic_obj(samples)
constrained_obj = apply_constraints(
obj=constrained_obj,
constraints=[feasible_con, infeasible_con],
samples=samples,
infeasible_cost=0.0,
)
self.assertTrue(torch.equal(obj(samples), constrained_obj))
# one feasible, one infeasible, infeasible_cost
obj = ConstrainedMCObjective(
objective=generic_obj,
constraints=[feasible_con, infeasible_con],
infeasible_cost=5.0,
)
samples = torch.randn(3, 2, device=device, dtype=dtype)
constrained_obj = generic_obj(samples)
constrained_obj = apply_constraints(
obj=constrained_obj,
constraints=[feasible_con, infeasible_con],
samples=samples,
infeasible_cost=5.0,
)
self.assertTrue(torch.equal(obj(samples), constrained_obj))
# one feasible, one infeasible, infeasible_cost, higher dimension
obj = ConstrainedMCObjective(
objective=generic_obj,
constraints=[feasible_con, infeasible_con],
infeasible_cost=5.0,
)
samples = torch.randn(4, 3, 2, device=device, dtype=dtype)
constrained_obj = generic_obj(samples)
constrained_obj = apply_constraints(
obj=constrained_obj,
constraints=[feasible_con, infeasible_con],
samples=samples,
infeasible_cost=5.0,
)
self.assertTrue(torch.equal(obj(samples), constrained_obj))
