File size: 9,089 Bytes
aa6c18e | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 | """OptimizationCampaign: manages the full lifecycle of an optimization campaign."""
import json
import time
from dataclasses import dataclass, field
from pathlib import Path
from typing import Callable, Dict, List, Optional, Tuple
import torch
from torch import Tensor
import pandas as pd
from physics_informed_bo.config import OptimizationConfig
from physics_informed_bo.experiment.designer import ExperimentDesigner
from physics_informed_bo.experiment.parameter_space import ParameterSpace
@dataclass
class ExperimentRecord:
"""Record of a single experiment."""
iteration: int
parameters: Dict[str, float]
objective: float
timestamp: float = field(default_factory=time.time)
metadata: Dict = field(default_factory=dict)
class OptimizationCampaign:
"""Manages an end-to-end Bayesian optimization campaign.
Provides:
- Full experiment tracking and history
- Save/load campaign state
- Convergence monitoring
- Human-in-the-loop workflow support
- Export to DataFrame for analysis
Example:
campaign = OptimizationCampaign(
name="polymer_optimization",
parameter_space=space,
physics_fn=my_physics_model,
config=OptimizationConfig(max_iterations=30),
)
# Automated loop
campaign.run_automated(objective_fn=evaluate_experiment)
# Or human-in-the-loop
next_exp = campaign.suggest_next()
# ... run experiment manually ...
campaign.report_result(next_exp, result_value)
"""
def __init__(
self,
name: str,
parameter_space: ParameterSpace,
physics_fn: Optional[Callable[[Tensor], Tensor]] = None,
initial_data: Optional[Tuple[Tensor, Tensor]] = None,
config: Optional[OptimizationConfig] = None,
maximize: bool = True,
):
self.name = name
self.maximize = maximize
self.config = config or OptimizationConfig()
self.parameter_space = parameter_space
self._designer = ExperimentDesigner(
parameter_space=parameter_space,
physics_fn=physics_fn,
initial_data=initial_data,
config=self.config,
)
self._history: List[ExperimentRecord] = []
self._iteration = 0
self._start_time = time.time()
# Track initial data if provided
if initial_data is not None:
X_init, y_init = initial_data
if y_init.dim() == 1:
y_init = y_init.unsqueeze(-1)
param_dicts = parameter_space.to_dict(X_init)
for params, y_val in zip(param_dicts, y_init):
self._history.append(
ExperimentRecord(
iteration=0,
parameters=params,
objective=float(y_val),
metadata={"source": "initial_data"},
)
)
def suggest_next(self, n: int = 1) -> List[Dict]:
"""Suggest the next experiment(s) to run.
Returns:
List of parameter dicts for suggested experiments.
"""
self._iteration += 1
candidates = self._designer.suggest(n)
return self.parameter_space.to_dict(candidates)
def report_result(
self,
parameters: Dict[str, float],
objective: float,
metadata: Optional[Dict] = None,
) -> None:
"""Report the result of a completed experiment.
Args:
parameters: The parameter values that were tested.
objective: The measured objective value.
metadata: Optional metadata about the experiment.
"""
record = ExperimentRecord(
iteration=self._iteration,
parameters=parameters,
objective=objective,
metadata=metadata or {},
)
self._history.append(record)
# Update the designer
X_new = self.parameter_space.from_dict(parameters).unsqueeze(0)
y_new = torch.tensor([[objective]], dtype=torch.float64)
self._designer.update(X_new, y_new)
def run_automated(
self,
objective_fn: Callable[[Dict[str, float]], float],
max_iterations: Optional[int] = None,
batch_size: int = 1,
callback: Optional[Callable] = None,
) -> pd.DataFrame:
"""Run a fully automated optimization loop.
Args:
objective_fn: Function that takes parameter dict and returns objective value.
max_iterations: Max iterations (defaults to config.max_iterations).
batch_size: Number of experiments per iteration.
callback: Optional callback(iteration, best_so_far) called each iteration.
Returns:
DataFrame of all experiments.
"""
max_iter = max_iterations or self.config.max_iterations
for i in range(max_iter):
# Suggest experiments
suggestions = self.suggest_next(batch_size)
# Evaluate
for params in suggestions:
objective = objective_fn(params)
self.report_result(params, objective)
# Callback
if callback:
best = self.get_best()
callback(i + 1, best)
# Check convergence
if self._check_convergence():
break
return self.to_dataframe()
def _check_convergence(self, window: int = 10, tolerance: float = 1e-4) -> bool:
"""Check if optimization has converged (no improvement in last `window` iterations)."""
if len(self._history) < window:
return False
recent = [r.objective for r in self._history[-window:]]
if self.maximize:
best_recent = max(recent)
best_before = max(r.objective for r in self._history[:-window])
return best_recent - best_before < tolerance
else:
best_recent = min(recent)
best_before = min(r.objective for r in self._history[:-window])
return best_before - best_recent < tolerance
def get_best(self) -> Dict:
"""Get the best experiment so far."""
if not self._history:
return {"parameters": {}, "objective": None}
if self.maximize:
best = max(self._history, key=lambda r: r.objective)
else:
best = min(self._history, key=lambda r: r.objective)
return {"parameters": best.parameters, "objective": best.objective}
def to_dataframe(self) -> pd.DataFrame:
"""Export campaign history as a pandas DataFrame."""
records = []
for r in self._history:
row = {"iteration": r.iteration, "objective": r.objective}
row.update(r.parameters)
row["timestamp"] = r.timestamp
records.append(row)
return pd.DataFrame(records)
def save(self, filepath: str) -> None:
"""Save campaign state to a JSON file."""
state = {
"name": self.name,
"maximize": self.maximize,
"iteration": self._iteration,
"history": [
{
"iteration": r.iteration,
"parameters": r.parameters,
"objective": r.objective,
"timestamp": r.timestamp,
"metadata": r.metadata,
}
for r in self._history
],
}
Path(filepath).write_text(json.dumps(state, indent=2))
def load(self, filepath: str) -> None:
"""Load campaign state from a JSON file."""
state = json.loads(Path(filepath).read_text())
self.name = state["name"]
self.maximize = state["maximize"]
self._iteration = state["iteration"]
self._history = [
ExperimentRecord(**r) for r in state["history"]
]
# Re-feed all data to the designer
if self._history:
all_params = [r.parameters for r in self._history]
X = torch.stack([self.parameter_space.from_dict(p) for p in all_params])
y = torch.tensor(
[r.objective for r in self._history], dtype=torch.float64
).unsqueeze(-1)
self._designer.update(X, y)
@property
def n_experiments(self) -> int:
return len(self._history)
def summary(self) -> Dict:
"""Campaign summary."""
best = self.get_best()
return {
"name": self.name,
"n_experiments": self.n_experiments,
"iteration": self._iteration,
"best": best,
"elapsed_time_s": time.time() - self._start_time,
"model_summary": self._designer.summary(),
}
|