sciarg: fix partitioning (#159)

* test label counts for labeled_partitions in sciarg

* fix and simplify tests

* fix allowing newlines between matching tags (important for abstract)

Author: ArneBinder

dataset_builders/pie/sciarg/sciarg.py

@@ -123,7 +123,8 @@ def _generate_document(self, example, **kwargs):
     def document_converters(self) -> DocumentConvertersType:
         regex_partitioner = RegexPartitioner(
             partition_layer_name="labeled_partitions",
-            pattern="<([^>/]+)>.*</\\1>",
+            # find matching tags, allow newlines in between (s flag) and capture the tag name
+            pattern="<([^>/]+)>(?s:.)*?</\\1>",
             label_group_id=1,
             label_whitelist=["Title", "Abstract", "H1"],
             skip_initial_partition=True,

tests/dataset_builders/pie/sciarg/test_sciarg.py

@@ -56,6 +56,21 @@
         "spans": {"background_claim": 2752, "data": 4093, "own_claim": 5450},
     },
 }
+CONVERTED_LAYER_MAPPING = {
+    "default": {
+        "spans": "labeled_spans",
+        "relations": "binary_relations",
+    },
+    "resolve_parts_of_same": {
+        "spans": "labeled_multi_spans",
+        "relations": "binary_relations",
+    },
+}
+FULL_LABEL_COUNTS_CONVERTED = {
+    variant: {CONVERTED_LAYER_MAPPING[variant][ln]: value for ln, value in counts.items()}
+    for variant, counts in FULL_LABEL_COUNTS.items()
+}
+LABELED_PARTITION_COUNTS = {"Abstract": 40, "H1": 340, "Title": 40}
 
 
 def resolve_annotation(annotation: Annotation) -> Any:
@@ -257,32 +272,25 @@ def converted_dataset(dataset, target_document_type) -> Optional[DatasetDict]:
     return dataset.to_document_type(target_document_type)
 
 
-def test_converted_datasets(converted_dataset, dataset_variant):
+def test_converted_datasets(converted_dataset, dataset_variant, target_document_type):
     if converted_dataset is not None:
         split_sizes = {name: len(ds) for name, ds in converted_dataset.items()}
         assert split_sizes == SPLIT_SIZES
         if dataset_variant == "default":
             expected_document_type = TextDocumentWithLabeledSpansAndBinaryRelations
-            layer_name_mapping = {
-                "spans": "labeled_spans",
-                "relations": "binary_relations",
-            }
         elif dataset_variant == "resolve_parts_of_same":
             expected_document_type = TextDocumentWithLabeledMultiSpansAndBinaryRelations
-            layer_name_mapping = {
-                "spans": "labeled_multi_spans",
-                "relations": "binary_relations",
-            }
         else:
             raise ValueError(f"Unknown dataset variant: {dataset_variant}")
 
+        assert issubclass(converted_dataset.document_type, expected_document_type)
         assert isinstance(converted_dataset["train"][0], expected_document_type)
 
         if TEST_FULL_DATASET:
-            expected_label_counts = {
-                layer_name_mapping[ln]: value
-                for ln, value in FULL_LABEL_COUNTS[dataset_variant].items()
-            }
+            # copy to avoid modifying the original dict
+            expected_label_counts = {**FULL_LABEL_COUNTS_CONVERTED[dataset_variant]}
+            if issubclass(target_document_type, TextDocumentWithLabeledPartitions):
+                expected_label_counts["labeled_partitions"] = LABELED_PARTITION_COUNTS
             assert_dataset_label_counts(converted_dataset, expected_label_counts)