Move predicate filtering out of datamodel

Part of the quest to remove the implementation details
of predicates out of DataModel and into the things that
actually care about them.

This slightly changes the behavior in the test because we don't
do any filtering either way, so we use ALL predicates from the
variable definitions

dedupe/datamodel.py

@@ -69,17 +69,12 @@ def _field_comparators(
             yield (var.field, comparator, start, stop)
             start = stop
 
-    def predicates(self, canopies: bool = True) -> set[Predicate]:
+    @property
+    def predicates(self) -> set[Predicate]:
         predicates = set()
         for var in self.primary_variables:
             for predicate in var.predicates:
-                if hasattr(predicate, "index"):
-                    is_canopy = hasattr(predicate, "canopy")
-                    if is_canopy == canopies:
-                        predicates.add(predicate)
-                else:
-                    predicates.add(predicate)
-
+                predicates.add(predicate)
         return predicates
 
     def distances(

dedupe/labeler.py

@@ -225,6 +225,20 @@ def _sample_indices(self, sample_size: int) -> Iterable[RecordIDPair]:
         return sample_ids
 
 
+def _filter_canopy_predicates(
+    predicates: Iterable[Predicate], canopies: bool
+) -> set[Predicate]:
+    result = set()
+    for predicate in predicates:
+        if hasattr(predicate, "index"):
+            is_canopy = hasattr(predicate, "canopy")
+            if is_canopy == canopies:
+                result.add(predicate)
+        else:
+            result.add(predicate)
+    return result
+
+
 class DedupeBlockLearner(BlockLearner):
     def __init__(
         self,
@@ -239,7 +253,8 @@ def __init__(
 
         index_data = sample_records(data, 50000)
         sampled_records = sample_records(index_data, N_SAMPLED_RECORDS)
-        preds = self.data_model.predicates()
+        preds = self.data_model.predicates
+        preds = _filter_canopy_predicates(preds, canopies=True)
 
         self.block_learner = training.DedupeBlockLearner(
             preds, sampled_records, index_data
@@ -293,7 +308,8 @@ def __init__(
         index_data = sample_records(data_2, 50000)
         sampled_records_2 = sample_records(index_data, N_SAMPLED_RECORDS)
 
-        preds = self.data_model.predicates(canopies=False)
+        preds = self.data_model.predicates
+        preds = _filter_canopy_predicates(preds, canopies=False)
 
         self.block_learner = training.RecordLinkBlockLearner(
             preds, sampled_records_1, sampled_records_2, index_data

tests/test_training.py

@@ -34,7 +34,7 @@ def setUp(self):
 
         self.block_learner = training.BlockLearner
         self.block_learner.blocker = dedupe.blocking.Fingerprinter(
-            self.data_model.predicates()
+            self.data_model.predicates
         )
         self.block_learner.blocker.index_all(
             {i: x for i, x in enumerate(self.training_records)}