File size: 7,487 Bytes
95c4c89
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
"""BoFire optimizer backend for physics-informed BO."""

from typing import Callable, Dict, List, Optional, Tuple

import torch
from torch import Tensor

from physics_informed_bo.config import OptimizationConfig
from physics_informed_bo.optimizers.base_optimizer import BaseOptimizer


class BoFireOptimizer(BaseOptimizer):
    """BoFire backend for chemistry/materials-focused Bayesian optimization.



    BoFire is designed for real-world experimental design in chemistry

    and materials science. It supports:

    - Complex parameter spaces (continuous, categorical, molecular)

    - Mixture constraints (sum-to-one)

    - Multi-objective optimization with Pareto fronts

    - Integration with domain-specific descriptors



    The physics model is incorporated as a prior mean function in BoFire's

    surrogate model specification.

    """

    def __init__(self, config: OptimizationConfig):
        super().__init__(config)
        self._domain = None
        self._strategy = None
        self._experiments_df = None

    def setup_domain(

        self,

        parameters: Dict[str, Dict],

        objectives: Dict[str, Dict],

        constraints: Optional[List[Dict]] = None,

    ) -> None:
        """Set up a BoFire domain with physics-informed features.



        Args:

            parameters: Dict of parameter specifications.

                Example: {"temp": {"type": "continuous", "bounds": (300, 500)}}

            objectives: Dict of objective specifications.

                Example: {"yield": {"type": "maximize", "weight": 1.0}}

            constraints: Optional list of constraint specifications.

                Example: [{"type": "linear", "features": ["x1", "x2"], "coeffs": [1, 1], "rhs": 1}]

        """
        try:
            from bofire.data_models.domain.api import Domain, Inputs, Outputs
            from bofire.data_models.features.api import (
                ContinuousInput,
                ContinuousOutput,
                CategoricalInput,
            )
            from bofire.data_models.objectives.api import MaximizeObjective, MinimizeObjective
            from bofire.data_models.constraints.api import (
                LinearInequalityConstraint,
                LinearEqualityConstraint,
            )
        except ImportError:
            raise ImportError(
                "BoFire is required for BoFireOptimizer. "
                "Install with: pip install bofire"
            )

        # Build input features
        input_features = []
        self._feature_names = []

        for name, spec in parameters.items():
            self._feature_names.append(name)
            if spec["type"] == "continuous":
                lb, ub = spec["bounds"]
                input_features.append(
                    ContinuousInput(key=name, bounds=(float(lb), float(ub)))
                )
            elif spec["type"] == "categorical":
                input_features.append(
                    CategoricalInput(key=name, categories=spec["categories"])
                )

        # Build output features
        output_features = []
        for name, spec in objectives.items():
            if spec.get("type", "maximize") == "maximize":
                obj = MaximizeObjective(w=spec.get("weight", 1.0))
            else:
                obj = MinimizeObjective(w=spec.get("weight", 1.0))
            output_features.append(ContinuousOutput(key=name, objective=obj))

        # Build constraints
        bofire_constraints = []
        if constraints:
            for c in constraints:
                if c["type"] == "linear_inequality":
                    bofire_constraints.append(
                        LinearInequalityConstraint(
                            features=c["features"],
                            coefficients=c["coeffs"],
                            rhs=c["rhs"],
                        )
                    )
                elif c["type"] == "linear_equality":
                    bofire_constraints.append(
                        LinearEqualityConstraint(
                            features=c["features"],
                            coefficients=c["coeffs"],
                            rhs=c["rhs"],
                        )
                    )

        self._domain = Domain(
            inputs=Inputs(features=input_features),
            outputs=Outputs(features=output_features),
            constraints=bofire_constraints if bofire_constraints else None,
        )

    def setup_strategy(self, strategy_type: str = "sobo") -> None:
        """Set up the BoFire optimization strategy.



        Args:

            strategy_type: One of 'sobo' (single-objective), 'mobo' (multi-objective),

                          'qehvi' (q-Expected Hypervolume Improvement).

        """
        try:
            from bofire.data_models.strategies.api import SoboStrategy, QehviStrategy
            from bofire.data_models.acquisition_functions.api import qEI, qNEI
            import bofire.strategies.api as strategies
        except ImportError:
            raise ImportError("BoFire is required. Install with: pip install bofire")

        if self._domain is None:
            raise RuntimeError("Call setup_domain() before setup_strategy().")

        if strategy_type == "sobo":
            strategy_data = SoboStrategy(domain=self._domain, acquisition_function=qEI())
        elif strategy_type in ("mobo", "qehvi"):
            strategy_data = QehviStrategy(domain=self._domain)
        else:
            raise ValueError(f"Unsupported strategy type: {strategy_type}")

        self._strategy = strategies.map(strategy_data)

    def suggest(

        self,

        n_candidates: int = 1,

        X_observed: Optional[Tensor] = None,

        y_observed: Optional[Tensor] = None,

    ) -> Tensor:
        """Suggest next experiments using BoFire."""
        if self._strategy is None:
            raise RuntimeError("Call setup_domain() and setup_strategy() first.")

        import pandas as pd

        # Tell strategy about existing experiments
        if self._experiments_df is not None:
            self._strategy.tell(self._experiments_df)

        candidates_df = self._strategy.ask(n_candidates)
        candidates = torch.tensor(
            candidates_df[self._feature_names].values, dtype=torch.float64
        )

        # Filter through physics constraints
        candidates = self._filter_feasible(candidates)
        return candidates[:n_candidates]

    def update(self, X_new: Tensor, y_new: Tensor) -> None:
        """Update BoFire with new observations."""
        import pandas as pd

        data = {}
        for i, name in enumerate(self._feature_names):
            data[name] = X_new[:, i].numpy()

        # Assume single objective for now
        output_keys = [f.key for f in self._domain.outputs.features]
        for i, key in enumerate(output_keys):
            if y_new.dim() > 1 and y_new.shape[1] > i:
                data[key] = y_new[:, i].numpy()
            else:
                data[key] = y_new.squeeze().numpy()

        new_df = pd.DataFrame(data)
        if self._experiments_df is None:
            self._experiments_df = new_df
        else:
            self._experiments_df = pd.concat(
                [self._experiments_df, new_df], ignore_index=True
            )