Update entropy search descriptions

Author: benmltu

botorch/acquisition/joint_entropy_search.py

@@ -31,6 +31,14 @@ class qLowerBoundJointEntropySearch(qLowerBoundMultiObjectiveJointEntropySearch)
     at a candidate point `X` and the optimal input-output pair.
 
     See [Tu2022]_ for a discussion on the estimation procedure.
+
+    NOTES:
+    (i) The estimated acquisition value could be negative.
+
+    (ii) The lower bound batch acquisition function might not be monotone in the
+    sense that adding more elements to the batch does not necessarily increase the
+    acquisition value. Specifically, the acquisition value can become smaller when
+    more inputs are added.
     """
 
     def __init__(

botorch/acquisition/multi_objective/joint_entropy_search.py

@@ -264,6 +264,14 @@ class qLowerBoundMultiObjectiveJointEntropySearch(
     at a candidate point `X` and the Pareto optimal input-output pairs.
 
     See [Tu2022]_ for a discussion on the estimation procedure.
+
+    NOTES:
+    (i) The estimated acquisition value could be negative.
+
+    (ii) The lower bound batch acquisition function might not be monotone in the
+    sense that adding more elements to the batch does not necessarily increase the
+    acquisition value. Specifically, the acquisition value can become smaller when
+    more inputs are added.
     """
 
     def __init__(

botorch/acquisition/multi_objective/max_value_entropy_search.py

@@ -200,6 +200,14 @@ class qLowerBoundMultiObjectiveMaxValueEntropySearch(
     at a candidate point `X` and the Pareto optimal outputs.
 
     See [Tu2022]_ for a discussion on the estimation procedure.
+
+    NOTES:
+    (i) The estimated acquisition value could be negative.
+
+    (ii) The lower bound batch acquisition function might not be monotone in the
+    sense that adding more elements to the batch does not necessarily increase the
+    acquisition value. Specifically, the acquisition value can become smaller when
+    more inputs are added.
     """
 
     def __init__(

botorch/acquisition/multi_objective/predictive_entropy_search.py

@@ -93,6 +93,8 @@ class qMultiObjectivePredictiveEntropySearch(AcquisitionFunction):
         iterations are required. Running too many EP iterations could lead to
         invertibility problems such as in problem (b). Setting a larger threshold
         or reducing the number of EP iterations could alleviate this.
+
+    (iii) The estimated acquisition value could be negative.
     """
 
     def __init__(

botorch/acquisition/predictive_entropy_search.py

@@ -31,6 +31,16 @@ class qPredictiveEntropySearch(qMultiObjectivePredictiveEntropySearch):
     This acquisition function approximates the mutual information between the
     observation at a candidate point `X` and the optimal set of inputs using
     expectation propagation (EP).
+
+    NOTES:
+    (i) The expectation propagation procedure can potentially fail due to the unstable
+    EP updates. This is however unlikely to happen in the single-objective setting
+    because we have much fewer EP factors. The jitter added in the training phase
+    (`ep_jitter`) and testing phase (`test_jitter`) can be increased to prevent
+    these failures from happening. More details in the description of
+    `qMultiObjectivePredictiveEntropySearch`.
+
+    (ii) The estimated acquisition value could be negative.
     """
 
     def __init__(
@@ -47,14 +57,6 @@ def __init__(
     ) -> None:
         r"""Predictive entropy search acquisition function.
 
-        NOTE:
-        The expectation propagation procedure can potentially fail due to the unstable
-        EP updates. This is however unlikely to happen in the single-objective setting
-        because we have much fewer EP factors. The jitter added in the training phase
-        (`ep_jitter`) and testing phase (`test_jitter`) can be increased to prevent
-        these failures from happening. More details in the description of
-        `qMultiObjectivePredictiveEntropySearch`.
-
         Args:
             model: A fitted single-outcome model.
             optimal_inputs: A `num_samples x d`-dim tensor containing the sampled