Log ei change

docs/examples/bayes_exp/bayesian_exploration.ipynb

@@ -33,10 +33,6 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {
-    "ExecuteTime": {
-     "end_time": "2023-09-19T18:54:51.865697700Z",
-     "start_time": "2023-09-19T18:54:48.783739700Z"
-    },
     "execution": {
      "iopub.execute_input": "2024-09-13T15:49:35.830514Z",
      "iopub.status.busy": "2024-09-13T15:49:35.830266Z",
@@ -99,10 +95,6 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {
-    "ExecuteTime": {
-     "end_time": "2023-09-19T18:54:51.916698600Z",
-     "start_time": "2023-09-19T18:54:51.867698800Z"
-    },
     "execution": {
      "iopub.execute_input": "2024-09-13T15:49:37.522962Z",
      "iopub.status.busy": "2024-09-13T15:49:37.522764Z",
@@ -122,10 +114,6 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {
-    "ExecuteTime": {
-     "end_time": "2023-09-19T18:54:55.212579300Z",
-     "start_time": "2023-09-19T18:54:51.897698400Z"
-    },
     "execution": {
      "iopub.execute_input": "2024-09-13T15:49:37.531732Z",
      "iopub.status.busy": "2024-09-13T15:49:37.531635Z",
@@ -147,10 +135,6 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {
-    "ExecuteTime": {
-     "end_time": "2023-09-19T18:54:55.227580Z",
-     "start_time": "2023-09-19T18:54:55.212579300Z"
-    },
     "execution": {
      "iopub.execute_input": "2024-09-13T15:49:39.056947Z",
      "iopub.status.busy": "2024-09-13T15:49:39.056826Z",
@@ -171,10 +155,6 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {
-    "ExecuteTime": {
-     "end_time": "2023-09-19T18:54:55.947579900Z",
-     "start_time": "2023-09-19T18:54:55.230580300Z"
-    },
     "execution": {
      "iopub.execute_input": "2024-09-13T15:49:39.062334Z",
      "iopub.status.busy": "2024-09-13T15:49:39.062246Z",
@@ -238,10 +218,6 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {
-    "ExecuteTime": {
-     "end_time": "2023-09-19T18:54:58.766007500Z",
-     "start_time": "2023-09-19T18:54:58.724008300Z"
-    },
     "execution": {
      "iopub.execute_input": "2024-09-13T15:49:40.599399Z",
      "iopub.status.busy": "2024-09-13T15:49:40.599290Z",
@@ -258,7 +234,7 @@
     "for name, val in X.generator.model.named_parameters():\n",
     "    print(f\"{name}:{val}\")\n",
     "\n",
-    "X.generator.model.models[2].covar_module.base_kernel.lengthscale"
+    "X.generator.model.models[2].covar_module.lengthscale"
    ]
   },
   {
@@ -279,10 +255,6 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {
-    "ExecuteTime": {
-     "end_time": "2023-09-19T18:54:58.776010Z",
-     "start_time": "2023-09-19T18:54:58.746008900Z"
-    },
     "execution": {
      "iopub.execute_input": "2024-09-13T15:49:40.605375Z",
      "iopub.status.busy": "2024-09-13T15:49:40.605292Z",
@@ -302,10 +274,6 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {
-    "ExecuteTime": {
-     "end_time": "2023-09-19T18:55:02.760921Z",
-     "start_time": "2023-09-19T18:54:58.766007500Z"
-    },
     "execution": {
      "iopub.execute_input": "2024-09-13T15:49:40.611135Z",
      "iopub.status.busy": "2024-09-13T15:49:40.611035Z",

docs/examples/bayes_exp/bayesian_exploration_w_interpolation.ipynb

@@ -264,7 +264,7 @@
     "for name, val in X.generator.model.named_parameters():\n",
     "    print(f\"{name}:{val}\")\n",
     "\n",
-    "X.generator.model.models[2].covar_module.base_kernel.lengthscale"
+    "X.generator.model.models[2].covar_module.lengthscale"
    ]
   },
   {

docs/examples/multi_objective_bayes_opt/mobo.ipynb

@@ -242,23 +242,6 @@
     "print(candidate[[\"x1\", \"x2\"]].to_numpy())\n",
     "ax.plot(*candidate[[\"x1\", \"x2\"]].to_numpy()[0], \"o\")"
    ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2024-09-13T15:57:37.628734Z",
-     "iopub.status.busy": "2024-09-13T15:57:37.628594Z",
-     "iopub.status.idle": "2024-09-13T15:57:37.935451Z",
-     "shell.execute_reply": "2024-09-13T15:57:37.935103Z"
-    }
-   },
-   "outputs": [],
-   "source": [
-    "%%time\n",
-    "candidate = X.generator.generate(1)"
-   ]
   }
  ],
  "metadata": {

xopt/generators/bayesian/bayesian_generator.py

@@ -97,10 +97,6 @@ class BayesianGenerator(Generator, ABC):
     computation_time : Optional[pd.DataFrame]
         A data frame tracking computation time in seconds.
 
-    log_transform_acquisition_function: Optional[bool]
-        Flag to determine if final acquisition function value should be
-        log-transformed before optimization.
-
     n_interpolate_samples: Optional[PositiveInt]
         Number of interpolation points to generate between last observation and next
         observation, requires n_candidates to be 1.
@@ -159,16 +155,11 @@ class BayesianGenerator(Generator, ABC):
         None,
         description="data frame tracking computation time in seconds",
     )
-    log_transform_acquisition_function: Optional[bool] = Field(
-        False,
-        description="flag to log transform the acquisition function before optimization",
-    )
     custom_objective: Optional[CustomXoptObjective] = Field(
         None,
         description="custom objective for optimization, replaces objective specified by VOCS",
     )
     n_interpolate_points: Optional[PositiveInt] = None
-    memory_length: Optional[PositiveInt] = None
 
     n_candidates: int = 1
 
@@ -507,6 +498,9 @@ def get_acquisition(self, model):
                 model, acq, self._get_constraint_callables(), sampler=sampler
             )
 
+            # log transform the result to handle the constraints
+            acq = LogAcquisitionFunction(acq)
+
         # apply fixed features if specified in the generator
         if self.fixed_features is not None:
             # get input dim
@@ -521,9 +515,6 @@ def get_acquisition(self, model):
                 acq_function=acq, d=dim, columns=columns, values=values
             )
 
-        if self.log_transform_acquisition_function:
-            acq = LogAcquisitionFunction(acq)
-
         return acq
 
     def get_optimum(self):

xopt/generators/bayesian/custom_botorch/log_acquisition_function.py

@@ -1,5 +1,4 @@
 from botorch.acquisition import AcquisitionFunction
-from botorch.utils import t_batch_mode_transform
 from botorch.utils.safe_math import log_softplus
 from torch import Tensor
 from torch.nn import Module
@@ -13,7 +12,6 @@ def __init__(
         Module.__init__(self)
         self.acq_func = acq_function
 
-    @t_batch_mode_transform(expected_q=1, assert_output_shape=False)
     def forward(self, X: Tensor) -> Tensor:
         # apply a softplus transform to avoid numerical gradient issues
         return log_softplus(self.acq_func(X), 1e-6)

xopt/generators/bayesian/expected_improvement.py

@@ -1,8 +1,8 @@
 import torch
 from botorch.acquisition import (
-    ExpectedImprovement,
-    qExpectedImprovement,
     ScalarizedPosteriorTransform,
+    LogExpectedImprovement,
+    qLogExpectedImprovement,
 )
 
 from xopt.generators.bayesian.bayesian_generator import (
@@ -30,7 +30,7 @@ def _get_acquisition(self, model):
         if self.n_candidates > 1 or isinstance(objective, CustomXoptObjective):
             # MC sampling for generating multiple candidate points
             sampler = self._get_sampler(model)
-            acq = qExpectedImprovement(
+            acq = qLogExpectedImprovement(
                 model,
                 best_f=best_f,
                 sampler=sampler,
@@ -42,7 +42,7 @@ def _get_acquisition(self, model):
             # note that the analytic version cannot handle custom objectives
             weights = set_botorch_weights(self.vocs).to(**self.tkwargs)
             posterior_transform = ScalarizedPosteriorTransform(weights)
-            acq = ExpectedImprovement(
+            acq = LogExpectedImprovement(
                 model, best_f=best_f, posterior_transform=posterior_transform
             )
 

xopt/generators/bayesian/mggpo.py

@@ -2,7 +2,9 @@
 
 import pandas as pd
 import torch
-from botorch.acquisition.multi_objective import qNoisyExpectedHypervolumeImprovement
+from botorch.acquisition.multi_objective.logei import (
+    qLogNoisyExpectedHypervolumeImprovement,
+)
 from pydantic import Field
 
 from xopt.generators.bayesian.objectives import create_mobo_objective
@@ -68,7 +70,7 @@ def _get_acquisition(self, model):
         inputs = self.get_input_data(self.data)
         sampler = self._get_sampler(model)
 
-        acq = qNoisyExpectedHypervolumeImprovement(
+        acq = qLogNoisyExpectedHypervolumeImprovement(
             model,
             X_baseline=inputs,
             prune_baseline=True,

xopt/generators/bayesian/mobo.py

@@ -2,7 +2,6 @@
 
 import torch
 from botorch.acquisition import FixedFeatureAcquisitionFunction
-from botorch.acquisition.multi_objective import qNoisyExpectedHypervolumeImprovement
 from botorch.acquisition.multi_objective.logei import (
     qLogNoisyExpectedHypervolumeImprovement,
 )
@@ -24,7 +23,7 @@ class MOBOGenerator(MultiObjectiveBayesianGenerator):
         description="flag to specify if pareto front points are to be used during "
         "optimization of the acquisition function",
     )
-    __doc__ = """Implements Multi-Objective Bayesian Optimization using the Expected
+    __doc__ = """Implements Multi-Objective Bayesian Optimization using the Log Expected
             Hypervolume Improvement acquisition function"""
 
     def _get_objective(self):
@@ -60,12 +59,7 @@ def _get_acquisition(self, model):
         inputs = self.get_input_data(self.data)
         sampler = self._get_sampler(model)
 
-        if self.log_transform_acquisition_function:
-            acqclass = qLogNoisyExpectedHypervolumeImprovement
-        else:
-            acqclass = qNoisyExpectedHypervolumeImprovement
-
-        acq = acqclass(
+        acq = qLogNoisyExpectedHypervolumeImprovement(
             model,
             X_baseline=inputs,
             constraints=self._get_constraint_callables(),

xopt/generators/bayesian/upper_confidence_bound.py

@@ -34,19 +34,12 @@ class UpperConfidenceBoundGenerator(BayesianGenerator):
     def validate_vocs_without_constraints(cls, v):
         if v.constraints:
             warnings.warn(
-                f"Using {cls.__name__} with constraints may lead to numerical issues if the base acquisition "
-                f"function has negative values."
+                f"Using {cls.__name__} with constraints will lead to invalid values "
+                f"if the base acquisition function has negative values. Use with "
+                f"caution."
             )
         return v
 
-    @field_validator("log_transform_acquisition_function")
-    def validate_log_transform_acquisition_function(cls, v):
-        if v:
-            raise ValueError(
-                "Log transform cannot be applied to potentially negative UCB "
-                "acquisition function."
-            )
-
     def _get_acquisition(self, model):
         objective = self._get_objective()
         if self.n_candidates > 1 or isinstance(objective, CustomXoptObjective):

xopt/tests/generators/bayesian/test_mobo.py

@@ -236,7 +236,6 @@ def test_log_mobo(self):
         gen = MOBOGenerator(
             vocs=tnk_vocs,
             reference_point=reference_point,
-            log_transform_acquisition_function=True,
         )
         gen = deepcopy(gen)
         gen.n_monte_carlo_samples = 20

xopt/tests/generators/bayesian/test_upper_confidence_bound.py

@@ -88,21 +88,6 @@ def test_in_xopt(self):
         for _ in range(1):
             X.step()
 
-    def test_positivity(self):
-        # for UCB to work properly with constraints, it must always be positive.
-        # to acheive this we set infeasible cost
-        ucb_gen = UpperConfidenceBoundGenerator(
-            vocs=TEST_VOCS_BASE,
-        )
-        ucb_gen.add_data(
-            pd.DataFrame({"x1": -1.0, "x2": -1.0, "y1": 100.0, "c1": -100}, index=[0])
-        )
-        ucb_gen.train_model()
-        # evaluate acqf
-        acqf = ucb_gen.get_acquisition(ucb_gen.model)
-        with torch.no_grad():
-            assert acqf(torch.tensor((-1.0, -1.0)).reshape(1, 1, 2)) >= 0.0
-
     def test_fixed_feature(self):
         # test with fixed feature not in vocs
         gen = UpperConfidenceBoundGenerator(vocs=TEST_VOCS_BASE)