Source code for botorch.models.cost

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

r"""
Cost models to be used with multi-fidelity optimization.
"""

from __future__ import annotations

from typing import Dict, Optional

import torch
from botorch.models.deterministic import DeterministicModel
from torch import Tensor


[docs]class AffineFidelityCostModel(DeterministicModel):
    r"""Affine cost model operating on fidelity parameters.

    For each (q-batch) element of a candidate set `X`, this module computes a
    cost of the form

        cost = fixed_cost + sum_j weights[j] * X[fidelity_dims[j]]
    """

    def __init__(
        self,
        fidelity_weights: Optional[Dict[int, float]] = None,
        fixed_cost: float = 0.01,
    ) -> None:
        r"""Affine cost model operating on fidelity parameters.

        Args:
            fidelity_weights: A dictionary mapping a subset of columns of `X`
                (the fidelity parameters) to it's associated weight in the
                affine cost expression. If omitted, assumes that the last
                column of X is the fidelity parameter with a weight of 1.0.
            fixed_cost: The fixed cost of running a single candidate point (i.e.
                an element of a q-batch).
        """
        if fidelity_weights is None:
            fidelity_weights = {-1: 1.0}
        super().__init__()
        self.fidelity_dims = sorted(fidelity_weights)
        self.fixed_cost = fixed_cost
        weights = torch.tensor([fidelity_weights[i] for i in self.fidelity_dims])
        self.register_buffer("weights", weights)
        self._num_outputs = 1

[docs]    def forward(self, X: Tensor) -> Tensor:
        r"""Evaluate the cost on a candidate set X.

        Computes a cost of the form

            cost = fixed_cost + sum_j weights[j] * X[fidelity_dims[j]]

        for each element of the q-batch

        Args:
            X: A `batch_shape x q x d'`-dim tensor of candidate points.

        Returns:
            A `batch_shape x q x 1`-dim tensor of costs.
        """
        # TODO: Consider different aggregation (i.e. max) across q-batch
        lin_cost = torch.einsum(
            "...f,f", X[..., self.fidelity_dims], self.weights.to(X)
        )
        return self.fixed_cost + lin_cost.unsqueeze(-1)