Merge pull request #59 from Leengit/Bald_to_BALD

STYLE: Use BALD instead of Bald, per arXiv:1906.08158

al_bench/factory.py

@@ -31,7 +31,7 @@ class ComputeCertainty:
     negative_entropy: str = "negative_entropy"
     batchbald: str = "batchbald"
     all_certainty_types: List[str]
-    all_certainty_types = ["confidence", "margin", "negative_entropy", "batchbald"]
+    all_certainty_types = [confidence, margin, negative_entropy, batchbald]
 
     def __init__(self, certainty_type, percentiles, cutoffs) -> None:
         """

al_bench/strategy.py

@@ -643,9 +643,9 @@ def select_next_indices(
         return predict_order
 
 
-class BaldStrategyHandler(GenericStrategyHandler):
+class BALDStrategyHandler(GenericStrategyHandler):
     def __init__(self) -> None:
-        super(BaldStrategyHandler, self).__init__()
+        super(BALDStrategyHandler, self).__init__()
 
     def select_next_indices(
         self,
@@ -654,17 +654,17 @@ def select_next_indices(
     ) -> NDArray[np.int_]:
         """
         Select new examples to be labeled by the expert.  This choses the unlabeled
-        examples based upon the BALD criterion.  (See also BatchBaldStrategyHandler.)
+        examples based upon the BALD criterion.  (See also BatchBALDStrategyHandler.)
         """
         print(f"self.predictions.shape = {self.predictions.shape}")
         raise NotImplementedError(
-            "BaldStrategyHandler::select_next_indices is not yet implemented."
+            "BALDStrategyHandler::select_next_indices is not yet implemented."
         )
 
 
-class BatchBaldStrategyHandler(GenericStrategyHandler):
+class BatchBALDStrategyHandler(GenericStrategyHandler):
     def __init__(self) -> None:
-        super(BatchBaldStrategyHandler, self).__init__()
+        super(BatchBALDStrategyHandler, self).__init__()
 
     def select_next_indices(
         self,
@@ -673,7 +673,7 @@ def select_next_indices(
     ) -> NDArray[np.int_]:
         """
         Select new examples to be labeled by the expert.  This chooses the unlabeled
-        examples based upon the Batch-BALD criterion.  (See also BaldStrategyHandler.)
+        examples based upon the BatchBALD criterion.  (See also BALDStrategyHandler.)
         """
         if validation_indices is None:
             validation_indices = np.array((), dtype=np.int64)
@@ -714,3 +714,8 @@ def select_next_indices(
                 dtype=torch.double,
             )
         return available_indices[candidates.indices]
+
+
+# Support legacy names for now
+BaldStrategyHandler = BALDStrategyHandler
+BatchBaldStrategyHandler = BatchBALDStrategyHandler

example/BayesianExample.ipynb

@@ -23,7 +23,7 @@
     "\n",
     "<p>Each ALBench simulation requires the user to specifiy a DatasetHandler, a ModelHandler, and a StrategyHandler.  A <b>DatasetHandler</b> provides an interface to the other handlers that is agnostic to the format of the dataset.  A <b>ModelHandler</b> provides an interface that is agnostic to such things as whether the model is implemented in TensorFlow or PyTorch.  A <b>StrategyHandler</b> implements a specific active learning strategy: for non-Bayesian models this includes random selection, least confidence, least margin, and maximum entropy; for Bayesian models this includes BatchBALD.\n",
     "    \n",
-    "The present implementation of the BatchBALD active learning strategy leans heavily upon use of the <a href=\"https://github.com/BlackHC/batchbald_redux/blob/master/README.md\">batchbald_redux Python package</a>.  Programmers who wish to add additional strategies to ALBench may benefit from looking at the implementation of the BatchBaldStrategyHandler within <a href=\"https://github.com/DigitalSlideArchive/ALBench/blob/main/al_bench/strategy.py\">al_bench/strategy.py</a>."
+    "The present implementation of the BatchBALD active learning strategy leans heavily upon use of the <a href=\"https://github.com/BlackHC/batchbald_redux/blob/master/README.md\">batchbald_redux Python package</a>.  Programmers who wish to add additional strategies to ALBench may benefit from looking at the implementation of the BatchBALDStrategyHandler within <a href=\"https://github.com/DigitalSlideArchive/ALBench/blob/main/al_bench/strategy.py\">al_bench/strategy.py</a>."
    ]
   },
   {
@@ -1754,7 +1754,7 @@
     "\n",
     "for name, my_strategy_handler in (\n",
     "    #   (\"BALD\", alb.strategy.BaldStrategyHandler()),\n",
-    "    (\"BatchBALD\", alb.strategy.BatchBaldStrategyHandler()),\n",
+    "    (\"BatchBALD\", alb.strategy.BatchBALDStrategyHandler()),\n",
     "):\n",
     "    print(f\"=== Begin Strategy {repr(name)} at {datetime.now()} ===\")\n",
     "    my_strategy_handler.set_dataset_handler(my_dataset_handler)\n",

test/test_0120_bayesian_model.py

@@ -143,7 +143,9 @@ def test_0120_bayesian_model() -> None:
     name: str
     my_strategy_handler: alb.strategy.AbstractStrategyHandler
     for name, my_strategy_handler in (
-        ("BatchBALD", alb.strategy.BatchBaldStrategyHandler()),
+        # ("BatchBALD", alb.strategy.BatchBALDStrategyHandler()),
+        # Deliberately use legacy name, to test that it is still supported.
+        ("BatchBald", alb.strategy.BatchBaldStrategyHandler()),
     ):
         my_strategy_handler.set_dataset_handler(my_dataset_handler)
         my_strategy_handler.set_model_handler(my_model_handler)