Source code for botorch.optim.parameter_constraints

#!/usr/bin/env python3
# Copyright (c) Meta Platforms, Inc. and affiliates.
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.

r"""
Utility functions for constrained optimization.
"""

from __future__ import annotations

from collections.abc import Callable
from functools import partial

import numpy as np
import numpy.typing as npt
import torch
from botorch.exceptions.errors import CandidateGenerationError, UnsupportedError
from botorch.optim.utils import columnwise_clamp, fix_features as apply_fix_features
from scipy.optimize import Bounds, minimize
from torch import Tensor


ScipyConstraintDict = dict[
    str, str | Callable[[np.ndarray], float] | Callable[[np.ndarray], np.ndarray]
]


[docs] def get_constraint_tolerance(dtype: torch.dtype) -> float: r"""Get the constraint tolerance for a given dtype. Args: dtype: The dtype to use. Returns: The constraint tolerance for the given dtype. """ if dtype == torch.double: return 1e-8 elif dtype == torch.float: return 1e-6 elif dtype == torch.half: return 1e-4 raise ValueError(f"Unsupported dtype {dtype}.")
[docs] def make_scipy_bounds( X: Tensor, lower_bounds: float | Tensor | None = None, upper_bounds: float | Tensor | None = None, ) -> Bounds | None: r"""Creates a scipy Bounds object for optimization Args: X: ``... x d`` tensor lower_bounds: Lower bounds on each column (last dimension) of ``X``. If this is a single float, then all columns have the same bound. upper_bounds: Lower bounds on each column (last dimension) of ``X``. If this is a single float, then all columns have the same bound. Returns: A scipy ``Bounds`` object if either lower_bounds or upper_bounds is not None, and None otherwise. Example: >>> X = torch.rand(5, 2) >>> scipy_bounds = make_scipy_bounds(X, 0.1, 0.8) """ if lower_bounds is None and upper_bounds is None: return None def _expand(bounds: float | Tensor, X: Tensor, lower: bool) -> Tensor: if bounds is None: ebounds = torch.full_like(X, float("-inf" if lower else "inf")) else: if not torch.is_tensor(bounds): bounds = torch.tensor(bounds) ebounds = bounds.expand_as(X) return _arrayify(ebounds).flatten() lb = _expand(bounds=lower_bounds, X=X, lower=True) ub = _expand(bounds=upper_bounds, X=X, lower=False) return Bounds(lb=lb, ub=ub, keep_feasible=True)
[docs] def make_scipy_linear_constraints( shapeX: torch.Size, inequality_constraints: list[tuple[Tensor, Tensor, float]] | None = None, equality_constraints: list[tuple[Tensor, Tensor, float]] | None = None, ) -> list[ScipyConstraintDict]: r"""Generate scipy constraints from torch representation. Args: shapeX: The shape of the torch.Tensor to optimize over (i.e. ``(b) x q x d``) inequality constraints: A list of tuples (indices, coefficients, rhs), with each tuple encoding an inequality constraint of the form ``\sum_i (X[indices[i]] * coefficients[i]) >= rhs``, where ``indices`` is a single-dimensional index tensor (long dtype) containing indices into the last dimension of ``X``, ``coefficients`` is a single-dimensional tensor of coefficients of the same length, and rhs is a scalar. equality constraints: A list of tuples (indices, coefficients, rhs), with each tuple encoding an inequality constraint of the form ``\sum_i (X[indices[i]] * coefficients[i]) == rhs`` (with ``indices`` and ``coefficients`` of the same form as in ``inequality_constraints``). Returns: A list of dictionaries containing callables for constraint function values and Jacobians and a string indicating the associated constraint type ("eq", "ineq"), as expected by ``scipy.optimize.minimize``. This function assumes that constraints are the same for each input batch, and broadcasts the constraints accordingly to the input batch shape. This function does support constraints across elements of a q-batch if the indices are a 2-d Tensor. Example: The following will enforce that ``x[1] + 0.5 x[3] >= -0.1`` for each ``x`` in both elements of the q-batch, and each of the 3 t-batches: >>> constraints = make_scipy_linear_constraints( >>> torch.Size([3, 2, 4]), >>> [(torch.tensor([1, 3]), torch.tensor([1.0, 0.5]), -0.1)], >>> ) The following will enforce that ``x[0, 1] + 0.5 x[1, 3] >= -0.1`` where x[0, :] is the first element of the q-batch and x[1, :] is the second element of the q-batch, for each of the 3 t-batches: >>> constraints = make_scipy_linear_constraints( >>> torch.size([3, 2, 4]) >>> [(torch.tensor([[0, 1], [1, 3]), torch.tensor([1.0, 0.5]), -0.1)], >>> ) """ constraints = [] if inequality_constraints is not None: for indcs, coeffs, rhs in inequality_constraints: constraints += _make_linear_constraints( indices=indcs, coefficients=coeffs, rhs=rhs, shapeX=shapeX, eq=False ) if equality_constraints is not None: for indcs, coeffs, rhs in equality_constraints: constraints += _make_linear_constraints( indices=indcs, coefficients=coeffs, rhs=rhs, shapeX=shapeX, eq=True ) return constraints
[docs] def eval_lin_constraint( x: npt.NDArray, flat_idxr: list[int], coeffs: npt.NDArray, rhs: float ) -> np.float64: r"""Evaluate a single linear constraint. Args: x: The input array. flat_idxr: The indices in ``x`` to consider. coeffs: The coefficients corresponding to the indices. rhs: The right-hand-side of the constraint. Returns: The evaluted constraint: ``\sum_i (coeffs[i] * x[i]) - rhs`` """ return np.sum(x[flat_idxr] * coeffs, -1) - rhs
[docs] def lin_constraint_jac( x: npt.NDArray, flat_idxr: list[int], coeffs: npt.NDArray, n: int ) -> npt.NDArray: r"""Return the Jacobian associated with a linear constraint. Args: x: The input array. flat_idxr: The indices for the elements of x that appear in the constraint. coeffs: The coefficients corresponding to the indices. n: number of elements Returns: The Jacobian. """ # TODO: Use sparse representation (not sure if scipy optim supports that) jac = np.zeros(n) jac[flat_idxr] = coeffs return jac
def _arrayify(X: Tensor) -> npt.NDArray: r"""Convert a torch.Tensor (any dtype or device) to a numpy (double) array. Args: X: The input tensor. Returns: A numpy array of double dtype with the same shape and data as ``X``. """ return X.cpu().detach().contiguous().double().clone().numpy() def _validate_linear_constraints_shape_input(shapeX: torch.Size) -> torch.Size: """ Validate ``shapeX`` input to ``_make_linear_constraints``. Check that it has either 2 or 3 dimensions, and add a scalar batch dimension if it is only 2d. """ if len(shapeX) not in (2, 3): raise UnsupportedError( f"`shapeX` must be `(b) x q x d` (at least two-dimensional). It is " f"{shapeX}." ) if len(shapeX) == 2: shapeX = torch.Size([1, *shapeX]) return shapeX def _validate_linear_constraints_indices_input(indices: Tensor, q: int, d: int) -> None: if indices.dim() > 2: raise UnsupportedError( "Linear constraints supported only on individual candidates and " "across q-batches, not across general batch shapes." ) elif indices.dim() == 2: if indices[:, 0].max() > q - 1: raise RuntimeError(f"Index out of bounds for {q}-batch") if indices[:, 1].max() > d - 1: raise RuntimeError(f"Index out of bounds for {d}-dim parameter tensor") elif indices.dim() == 1: if indices.max() > d - 1: raise RuntimeError(f"Index out of bounds for {d}-dim parameter tensor") else: raise ValueError("`indices` must be at least one-dimensional") def _make_linear_constraints( indices: Tensor, coefficients: Tensor, rhs: float, shapeX: torch.Size, eq: bool = False, ) -> list[ScipyConstraintDict]: r"""Create linear constraints to be used by ``scipy.optimize.minimize``. Encodes constraints of the form ``\sum_i (coefficients[i] * X[..., indices[i]]) ? rhs`` where ``?`` can be designated either as ``>=`` by setting ``eq=False``, or as ``=`` by setting ``eq=True``. If indices is one-dimensional, the constraints are broadcasted across all elements of the q-batch. If indices is two-dimensional, then constraints are applied across elements of a q-batch. In either case, constraints are created for all t-batches. Args: indices: A tensor of shape ``c`` or ``c x 2``, where c is the number of terms in the constraint. If single-dimensional, contains the indices of the dimensions of the feature space that occur in the linear constraint. If two-dimensional, contains pairs of indices of the q-batch (0) and the feature space (1) that occur in the linear constraint. coefficients: A single-dimensional tensor of coefficients with the same number of elements as ``indices``. rhs: The right hand side of the constraint. shapeX: The shape of the torch tensor to construct the constraints for (i.e. ``(b) x q x d``). Must have two or three dimensions. eq: If True, return an equality constraint, o/w return an inequality constraint (indicated by "eq" / "ineq" value of the ``type`` key). Returns: A list of constraint dictionaries with the following keys - "type": Indicates the type of the constraint ("eq" if ``eq=True``, "ineq" o/w) - "fun": A callable evaluating the constraint value on ``x``, a flattened version of the input tensor ``X``, returning a scalar. - "jac": A callable evaluating the constraint's Jacobian on ``x``, a flattened version of the input tensor ``X``, returning a numpy array. >>> shapeX = torch.Size([3, 5, 4]) >>> constraints = _make_linear_constraints( ... indices=torch.tensor([1., 2.]), ... coefficients=torch.tensor([-0.5, 1.3]), ... rhs=0.49, ... shapeX=shapeX, ... eq=True ... ) >>> len(constraints) 15 >>> constraints[0].keys() dict_keys(['type', 'fun', 'jac']) >>> x = np.arange(60).reshape(shapeX) >>> constraints[0]["fun"](x) 1.61 # 1 * -0.5 + 2 * 1.3 - 0.49 >>> constraints[0]["jac"](x) [0., -0.5, 1.3, 0., 0., ...] >>> constraints[1]["fun"](x) # 4.81 """ shapeX = _validate_linear_constraints_shape_input(shapeX) b, q, d = shapeX _validate_linear_constraints_indices_input(indices, q, d) n = shapeX.numel() constraints: list[ScipyConstraintDict] = [] coeffs = _arrayify(coefficients) ctype = "eq" if eq else "ineq" offsets = [q * d, d] if indices.dim() == 2: # indices has two dimensions (potential constraints across q-batch elements) # rule is [i, j, k] is at # i * offsets[0] + j * offsets[1] + k for i in range(b): list_ind = (idx.tolist() for idx in indices) idxr = [i * offsets[0] + idx[0] * offsets[1] + idx[1] for idx in list_ind] fun = partial( eval_lin_constraint, flat_idxr=idxr, coeffs=coeffs, rhs=float(rhs) ) jac = partial(lin_constraint_jac, flat_idxr=idxr, coeffs=coeffs, n=n) constraints.append({"type": ctype, "fun": fun, "jac": jac}) elif indices.dim() == 1: # indices is one-dim - broadcast constraints across q-batches and t-batches for i in range(b): for j in range(q): idxr = (i * offsets[0] + j * offsets[1] + indices).tolist() fun = partial( eval_lin_constraint, flat_idxr=idxr, coeffs=coeffs, rhs=float(rhs) ) jac = partial(lin_constraint_jac, flat_idxr=idxr, coeffs=coeffs, n=n) constraints.append({"type": ctype, "fun": fun, "jac": jac}) return constraints def _make_nonlinear_constraints( f_np_wrapper: Callable, nlc: Callable, is_intrapoint: bool, shapeX: torch.Size ) -> list[ScipyConstraintDict]: """Create nonlinear constraints to be used by ``scipy.optimize.minimize``. Args: f_np_wrapper: A wrapper function that given a constraint evaluates the value and gradient (using autograd) of a numpy input and returns both the objective and the gradient. nlc: Callable representing a constraint of the form ``callable(x) >= 0``. In case of an intra-point constraint, ``callable()``takes in an one-dimensional tensor of shape ``d`` and returns a scalar. In case of an inter-point constraint, ``callable()`` takes a two dimensional tensor of shape ``q x d`` and again returns a scalar. is_intrapoint: A Boolean indicating if a constraint is an intra-point or inter-point constraint (see the docstring of the ``inequality_constraints`` argument to ``optimize_acqf()``). shapeX: Shape of the three-dimensional batch X, that should be optimized. Returns: A list of constraint dictionaries with the following keys - "type": Indicates the type of the constraint, here always "ineq". - "fun": A callable evaluating the constraint value on ``x``, a flattened version of the input tensor ``X``, returning a scalar. - "jac": A callable evaluating the constraint's Jacobian on ``x``, a flattened version of the input tensor ``X``, returning a numpy array. """ shapeX = _validate_linear_constraints_shape_input(shapeX) b, q, _ = shapeX constraints = [] def get_intrapoint_constraint(b: int, q: int, nlc: Callable) -> Callable: return lambda x: nlc(x[b, q]) def get_interpoint_constraint(b: int, nlc: Callable) -> Callable: return lambda x: nlc(x[b]) if is_intrapoint: for i in range(b): for j in range(q): f_obj, f_grad = _make_f_and_grad_nonlinear_inequality_constraints( f_np_wrapper=f_np_wrapper, nlc=get_intrapoint_constraint(b=i, q=j, nlc=nlc), ) constraints.append({"type": "ineq", "fun": f_obj, "jac": f_grad}) else: for i in range(b): f_obj, f_grad = _make_f_and_grad_nonlinear_inequality_constraints( f_np_wrapper=f_np_wrapper, nlc=get_interpoint_constraint(b=i, nlc=nlc), ) constraints.append({"type": "ineq", "fun": f_obj, "jac": f_grad}) return constraints def _generate_unfixed_nonlin_constraints( constraints: list[tuple[Callable[[Tensor], Tensor], bool]] | None, fixed_features: dict[int, float], dimension: int, ) -> list[Callable[[Tensor], Tensor]] | None: """Given a dictionary of fixed features, returns a list of callables for nonlinear inequality constraints expecting only a tensor with the non-fixed features as input. """ if not constraints: return constraints selector = [] idx_X, idx_f = 0, dimension - len(fixed_features) for i in range(dimension): if i in fixed_features.keys(): selector.append(idx_f) idx_f += 1 else: selector.append(idx_X) idx_X += 1 values = torch.tensor(list(fixed_features.values()), dtype=torch.double) def _wrap_nonlin_constraint( constraint: Callable[[Tensor], Tensor], ) -> Callable[[Tensor], Tensor]: def new_nonlin_constraint(X: Tensor) -> Tensor: ivalues = values.to(X).expand(*X.shape[:-1], len(fixed_features)) X_perm = torch.cat([X, ivalues], dim=-1) return constraint(X_perm[..., selector]) return new_nonlin_constraint return [ (_wrap_nonlin_constraint(constraint=nlc), is_intrapoint) for nlc, is_intrapoint in constraints ] def _generate_unfixed_lin_constraints( constraints: list[tuple[Tensor, Tensor, float]] | None, fixed_features: dict[int, float], dimension: int, eq: bool, ) -> list[tuple[Tensor, Tensor, float]] | None: # If constraints is None or an empty list, then return itself if not constraints: return constraints # replace_index generates the new indices for the unfixed dimensions # after eliminating the fixed dimensions. # Example: dimension = 5, ff.keys() = [1, 3], replace_index = {0: 0, 2: 1, 4: 2} unfixed_keys = sorted(set(range(dimension)) - set(fixed_features)) unfixed_keys = torch.tensor(unfixed_keys).to(constraints[0][0]) replace_index = torch.arange(dimension - len(fixed_features)).to(constraints[0][0]) new_constraints = [] # parse constraints one-by-one for constraint_id, (indices, coefficients, rhs) in enumerate(constraints): new_rhs = rhs new_indices = [] new_coefficients = [] # the following unsqueeze is done to facilitate a simpler for-loop. indices_2dim = indices if indices.ndim == 2 else indices.unsqueeze(-1) for coefficient, index in zip(coefficients, indices_2dim): ffval_or_None = fixed_features.get(index[-1].item()) # if ffval_or_None is None, then the index is not fixed if ffval_or_None is None: new_indices.append(index) new_coefficients.append(coefficient) # otherwise, we "remove" the constraints corresponding to that index else: new_rhs = new_rhs - coefficient.item() * ffval_or_None # all indices were fixed, so the constraint is gone. if len(new_indices) == 0: if (eq and new_rhs != 0) or (not eq and new_rhs > 0): prefix = "Eq" if eq else "Ineq" raise CandidateGenerationError( f"{prefix}uality constraint {constraint_id} not met " "with fixed_features." ) else: # However, one key transformation has to be noted. # new_indices is with respect to the older (fuller) domain, and so it will # have to be converted using replace_index. new_indices = torch.stack(new_indices, dim=0) # generate new index location after the removal of fixed_features indices new_indices_dim_d = new_indices[:, -1].unsqueeze(-1) new_indices_dim_d = replace_index[ torch.nonzero(new_indices_dim_d == unfixed_keys, as_tuple=True)[1] ] new_indices[:, -1] = new_indices_dim_d # squeeze(-1) is a no-op if dim -1 is not singleton new_indices.squeeze_(-1) # convert new_coefficients to Tensor new_coefficients = torch.stack(new_coefficients) new_constraints.append((new_indices, new_coefficients, new_rhs)) return new_constraints def _make_f_and_grad_nonlinear_inequality_constraints( f_np_wrapper: Callable, nlc: Callable ) -> tuple[Callable[[Tensor], Tensor], Callable[[Tensor], Tensor]]: """ Create callables for objective + grad for the nonlinear inequality constraints. The Scipy interface requires specifying separate callables and we use caching to avoid evaluating the same input twice. This caching only works if the returned functions are evaluated on the same input in immediate sequence (i.e., calling ``f_obj(X_1)``, ``f_grad(X_1)`` will result in a single forward pass, while ``f_obj(X_1)``, ``f_grad(X_2)``, ``f_obj(X_1)`` will result in three forward passes). """ def f_obj_and_grad(x): obj, grad = f_np_wrapper(x, f=nlc) return obj, grad cache = {"X": None, "obj": None, "grad": None} def f_obj(X): X_c = cache["X"] if X_c is None or not np.array_equal(X_c, X): cache["X"] = X.copy() cache["obj"], cache["grad"] = f_obj_and_grad(X) return cache["obj"] def f_grad(X): X_c = cache["X"] if X_c is None or not np.array_equal(X_c, X): cache["X"] = X.copy() cache["obj"], cache["grad"] = f_obj_and_grad(X) return cache["grad"] return f_obj, f_grad
[docs] def project_to_equality_constraints( X: Tensor, equality_constraints: list[tuple[Tensor, Tensor, float]], ) -> Tensor: r"""Project X onto the equality constraint manifold via least-squares. For linear equality constraints of the form ``Ax = b``, this finds the closest point to X (in L2 sense) that satisfies all constraints, using the closed-form least-squares projection: ``X_proj = X + A^T (A A^T)^{-1} (b - A X)``. This operates on each point in the q-batch independently (intra-point constraints only). Args: X: A ``... x q x d``-dim tensor of inputs. equality_constraints: A list of tuples (indices, coefficients, rhs), with each tuple encoding an equality constraint of the form ``sum_i (X[indices[i]] * coefficients[i]) = rhs``. Only supports 1-d indices (intra-point constraints). Returns: A tensor of the same shape as X, projected onto the constraint manifold. """ if not equality_constraints: return X # Filter to intra-point constraints only (1-d indices). intra_constraints = [ (indices, coefficients, rhs) for indices, coefficients, rhs in equality_constraints if indices.ndim <= 1 ] if not intra_constraints: return X d = X.shape[-1] n_constraints = len(intra_constraints) # Build constraint matrix A and rhs vector b A = torch.zeros(n_constraints, d, dtype=X.dtype, device=X.device) b = torch.zeros(n_constraints, dtype=X.dtype, device=X.device) for i, (indices, coefficients, rhs) in enumerate(intra_constraints): A[i, indices.long()] = coefficients.to(dtype=X.dtype, device=X.device) b[i] = rhs # Compute residual = b - A @ x for each point. # A: (n_constraints, d), X: (... x q x d) # residual: (... x q x n_constraints) residual = b - torch.einsum("cd,...qd->...qc", A, X) # Compute correction = A^T @ (A A^T)^{-1} @ residual AAT = A @ A.T # (n_constraints, n_constraints) # Solve AAT @ lam = residual^T for lam # lam: (... x q x n_constraints) lam = torch.linalg.solve(AAT, residual.unsqueeze(-1)).squeeze(-1) # correction = A^T @ lam, i.e., sum over constraints # A.T: (d, n_constraints), lam: (... x q x n_constraints) correction = torch.einsum("dc,...qc->...qd", A.T, lam) return X + correction
[docs] def nonlinear_constraint_is_feasible( nonlinear_inequality_constraint: Callable, is_intrapoint: bool, x: Tensor, tolerance: float | None = None, ) -> Tensor: """Checks if a nonlinear inequality constraint is fulfilled (within tolerance). Args: nonlinear_inequality_constraint: Callable to evaluate the constraint. intra: If True, the constraint is an intra-point constraint that is applied pointwise and is broadcasted over the q-batch. Else, the constraint has to evaluated over the whole q-batch and is a an inter-point constraint. x: Tensor of shape (batch x q x d). tolerance: Rather than using the exact ``const(x) >= 0`` constraint, this helper checks feasibility of ``const(x) >= -tolerance``. This avoids marking the candidates as infeasible due to tiny violations. Returns: A boolean tensor of shape (batch) indicating if the constraint is satified by the corresponding batch of ``x``. """ if tolerance is None: tolerance = get_constraint_tolerance(dtype=x.dtype) def check_x(x: Tensor) -> bool: return _arrayify(nonlinear_inequality_constraint(x)).item() >= -tolerance x_flat = x.view(-1, *x.shape[-2:]) is_feasible = torch.ones(x_flat.shape[0], dtype=torch.bool, device=x.device) for i, x_ in enumerate(x_flat): if is_intrapoint: is_feasible[i] &= all(check_x(x__) for x__ in x_) else: is_feasible[i] &= check_x(x_) return is_feasible.view(x.shape[:-2])
[docs] def make_scipy_nonlinear_inequality_constraints( nonlinear_inequality_constraints: list[tuple[Callable, bool]], f_np_wrapper: Callable, x0: Tensor, shapeX: torch.Size, ) -> list[dict]: r"""Generate Scipy nonlinear inequality constraints from callables. Args: nonlinear_inequality_constraints: A list of tuples representing the nonlinear inequality constraints. The first element in the tuple is a callable representing a constraint of the form ``callable(x) >= 0``. In case of an intra-point constraint, ``callable()``takes in an one-dimensional tensor of shape ``d`` and returns a scalar. In case of an inter-point constraint, ``callable()`` takes a two dimensional tensor of shape ``q x d`` and again returns a scalar. The second element is a boolean, indicating if it is an intra-point or inter-point constraint (``True`` for intra-point. ``False`` for inter-point). For more information on intra-point vs inter-point constraints, see the docstring of the ``inequality_constraints`` argument to ``optimize_acqf()``. The constraints will later be passed to the scipy solver. f_np_wrapper: A wrapper function that given a constraint evaluates the value and gradient (using autograd) of a numpy input and returns both the objective and the gradient. x0: The starting point for SLSQP. We return this starting point in (rare) cases where SLSQP fails and thus require it to be feasible. shapeX: Shape of the three-dimensional batch X, that should be optimized. Returns: A list of dictionaries containing callables for constraint function values and Jacobians and a string indicating the associated constraint type ("eq", "ineq"), as expected by ``scipy.optimize.minimize``. """ scipy_nonlinear_inequality_constraints = [] for constraint in nonlinear_inequality_constraints: if not isinstance(constraint, tuple): raise ValueError( f"A nonlinear constraint has to be a tuple, got {type(constraint)}." ) if len(constraint) != 2: raise ValueError( "A nonlinear constraint has to be a tuple of length 2, " f"got length {len(constraint)}." ) nlc, is_intrapoint = constraint if not nonlinear_constraint_is_feasible( nlc, is_intrapoint=is_intrapoint, x=x0.reshape(shapeX) ).all(): raise ValueError( "`batch_initial_conditions` must satisfy the non-linear inequality " "constraints." ) scipy_nonlinear_inequality_constraints += _make_nonlinear_constraints( f_np_wrapper=f_np_wrapper, nlc=nlc, is_intrapoint=is_intrapoint, shapeX=shapeX, ) return scipy_nonlinear_inequality_constraints
[docs] def evaluate_feasibility( X: Tensor, inequality_constraints: list[tuple[Tensor, Tensor, float]] | None = None, equality_constraints: list[tuple[Tensor, Tensor, float]] | None = None, nonlinear_inequality_constraints: list[tuple[Callable, bool]] | None = None, tolerance: float | None = None, ) -> Tensor: r"""Evaluate feasibility of candidate points (within a tolerance). Args: X: The candidate tensor of shape ``batch x q x d``. inequality_constraints: A list of tuples (indices, coefficients, rhs), with each tuple encoding an inequality constraint of the form ``\sum_i (X[indices[i]] * coefficients[i]) >= rhs``. ``indices`` and ``coefficients`` should be torch tensors. See the docstring of ``make_scipy_linear_constraints`` for an example. When q=1, or when applying the same constraint to each candidate in the batch (intra-point constraint), ``indices`` should be a 1-d tensor. For inter-point constraints, in which the constraint is applied to the whole batch of candidates, ``indices`` must be a 2-d tensor, where in each row ``indices[i] =(k_i, l_i)`` the first index ``k_i`` corresponds to the ``k_i``-th element of the ``q``-batch and the second index ``l_i`` corresponds to the ``l_i``-th feature of that element. equality_constraints: A list of tuples (indices, coefficients, rhs), with each tuple encoding an equality constraint of the form ``\sum_i (X[indices[i]] * coefficients[i]) = rhs``. See the docstring of ``make_scipy_linear_constraints`` for an example. nonlinear_inequality_constraints: A list of tuples representing the nonlinear inequality constraints. The first element in the tuple is a callable representing a constraint of the form ``callable(x) >= 0``. In case of an intra-point constraint, ``callable()``takes in an one-dimensional tensor of shape ``d`` and returns a scalar. In case of an inter-point constraint, ``callable()`` takes a two dimensional tensor of shape ``q x d`` and again returns a scalar. The second element is a boolean, indicating if it is an intra-point or inter-point constraint (``True`` for intra-point. ``False`` for inter-point). For more information on intra-point vs inter-point constraints, see the docstring of the ``inequality_constraints`` argument. tolerance: The tolerance used to check the feasibility of constraints. For inequality constraints, we check if ``const(X) >= rhs - tolerance``. For equality constraints, we check if ``abs(const(X) - rhs) < tolerance``. For non-linear inequality constraints, we check if ``const(X) >= -tolerance``. This avoids marking the candidates as infeasible due to tiny violations. Returns: A boolean tensor of shape ``batch`` indicating if the corresponding candidate of shape ``q x d`` is feasible. """ if tolerance is None: tolerance = get_constraint_tolerance(dtype=X.dtype) is_feasible = torch.ones(X.shape[:-2], device=X.device, dtype=torch.bool) if inequality_constraints is not None: for idx, coef, rhs in inequality_constraints: if idx.ndim == 1: # Intra-point constraints. is_feasible &= ( (X[..., idx] * coef).sum(dim=-1) >= rhs - tolerance ).all(dim=-1) else: # Inter-point constraints. is_feasible &= (X[..., idx[:, 0], idx[:, 1]] * coef).sum( dim=-1 ) >= rhs - tolerance if equality_constraints is not None: for idx, coef, rhs in equality_constraints: if idx.ndim == 1: # Intra-point constraints. is_feasible &= ( ((X[..., idx] * coef).sum(dim=-1) - rhs).abs() < tolerance ).all(dim=-1) else: # Inter-point constraints. is_feasible &= ( (X[..., idx[:, 0], idx[:, 1]] * coef).sum(dim=-1) - rhs ).abs() < tolerance if nonlinear_inequality_constraints is not None: for const, intra in nonlinear_inequality_constraints: is_feasible &= nonlinear_constraint_is_feasible( nonlinear_inequality_constraint=const, is_intrapoint=intra, x=X, tolerance=tolerance, ) return is_feasible
[docs] def project_to_feasible_space_via_slsqp( X: Tensor, bounds: Tensor, inequality_constraints: list[tuple[Tensor, Tensor, float]] | None = None, equality_constraints: list[tuple[Tensor, Tensor, float]] | None = None, fixed_features: dict[int, float | Tensor] | None = None, ) -> Tensor: """Project X onto the feasible space by solving a quadratic program. This uses SLSQP with gradients to solve the quadratic program. NOTE: A proper specialized QP solver would be a better choice here, but we'd like to avoid adding dependency on additional packages. SLSQP should be able to solve this reliably and quickly since the dimension is typically low and the number of constraints is typically limited. Args: X: A ``(batch_shape x) n x d``-dim tensor of inputs. bounds: A ``2 x d``-dim tensor of lower and upper bounds. inequality_constraints: A list of tuples (indices, coefficients, rhs), with each tuple encoding an inequality constraint of the form ``sum_i (X[indices[i]] * coefficients[i]) >= rhs``. ``indices`` and ``coefficients`` should be torch tensors. See the docstring of ``make_scipy_linear_constraints`` for an example. equality_constraints: A list of tuples (indices, coefficients, rhs). fixed_features: A dictionary mapping feature indices to their fixed values. These dimensions will not be modified during projection. Values can be scalars (applied to all elements) or 1D tensors matching the batch size of X (for per-element fixed values). Returns: A ``(batch_shape x) n x d``-dim tensor of projected values. """ if inequality_constraints is None and equality_constraints is None: return X d = X.shape[-1] lb = _arrayify(bounds[0].expand_as(X)).flatten() ub = _arrayify(bounds[1].expand_as(X)).flatten() # If there are fixed features, constrain those dimensions by setting their # bounds to equal the current value. This prevents the optimizer from # modifying them during projection. We use fix_features to apply the fixed # values to X, then extract the values for setting the bounds. if fixed_features: X_fixed = apply_fix_features(X, fixed_features, replace_current_value=True) # Set bounds for fixed dimensions to match the fixed values X_fixed_flat = _arrayify(X_fixed).flatten() for idx in fixed_features.keys(): # For each row in the flattened structure, set bounds at dimension idx n_rows = X.numel() // d for i in range(n_rows): flat_idx = i * d + idx lb[flat_idx] = X_fixed_flat[flat_idx] ub[flat_idx] = X_fixed_flat[flat_idx] bounds_scipy = Bounds(lb=lb, ub=ub, keep_feasible=True) constraints = make_scipy_linear_constraints( shapeX=X.shape, inequality_constraints=inequality_constraints, equality_constraints=equality_constraints, ) # Define squared distance objective X_np = X.flatten().detach().cpu().numpy() def objective(x: np.ndarray): return 0.5 * np.sum((x - X_np) ** 2) def grad_objective(x: np.ndarray): return x - X_np x0 = ( columnwise_clamp(X=X, lower=bounds[0], upper=bounds[1], raise_on_violation=True) .detach() .cpu() .numpy() .flatten() ) # NOTE: A proper specialized QP solver would be a better choice here, # but we'd like to avoid adding dependency on additional packages. # SLSQP should be able to solve this reliably and quickly since the # dimension is typically low and the number of constraints is typically # limited. result = minimize( fun=objective, x0=x0, method="SLSQP", jac=grad_objective, bounds=bounds_scipy, constraints=constraints, tol=get_constraint_tolerance(dtype=X.dtype), ) if not result.success: raise CandidateGenerationError(f"Optimization failed: {result.message}") return torch.from_numpy(result.x).to(X).view(X.shape)