feat: Add KeyValueEvaluator (#140)

* add KeyValueEvaluator

Signed-off-by: Saidgurbuz <[email protected]>

* add test script for keyvalue_evaluator

Signed-off-by: Saidgurbuz <[email protected]>

* add keyvalue_evaluator in cli

Signed-off-by: Saidgurbuz <[email protected]>

* add entity and link difference metrics in keyvalue evaluator

Signed-off-by: Saidgurbuz <[email protected]>

* update dependencies and fix style

Signed-off-by: Saidgurbuz <[email protected]>

* Refactor to remove levenshtein dependency

Signed-off-by: Christoph Auer <[email protected]>

---------

Signed-off-by: Saidgurbuz <[email protected]>
Signed-off-by: Christoph Auer <[email protected]>
Co-authored-by: Christoph Auer <[email protected]>

Author: Saidgurbuz

docling_eval/cli/main.py

@@ -55,6 +55,10 @@
 from docling_eval.evaluators.base_evaluator import DatasetEvaluationType
 from docling_eval.evaluators.bbox_text_evaluator import BboxTextEvaluator
 from docling_eval.evaluators.doc_structure_evaluator import DocStructureEvaluator
+from docling_eval.evaluators.keyvalue_evaluator import (
+    DatasetKeyValueEvaluation,
+    KeyValueEvaluator,
+)
 from docling_eval.evaluators.layout_evaluator import (
     DatasetLayoutEvaluation,
     LabelFilteringStrategy,
@@ -570,7 +574,21 @@ def evaluate(
             idir,
             split=split,
         )
+        with open(save_fn, "w") as fd:
+            json.dump(
+                evaluation.model_dump(),
+                fd,
+                indent=2,
+                sort_keys=True,
+                ensure_ascii=False,
+            )
 
+    elif modality == EvaluationModality.KEY_VALUE:
+        keyvalue_evaluator = KeyValueEvaluator()
+        evaluation = keyvalue_evaluator(  # type: ignore
+            idir,
+            split=split,
+        )
         with open(save_fn, "w") as fd:
             json.dump(
                 evaluation.model_dump(),