change core recipe test; remove unused vila test

tests/test_predictors/test_vila_predictor.py

@@ -82,17 +82,3 @@ def test_vila_predictors(self):
             "allenai/ivila-row-layoutlm-finetuned-s2vl-v2"
         )
         results_with_rows = predictor_with_rows.predict(doc, subpage_per_run=2)
-
-    def test_vila_predictors_with_special_unicode_inputs(self):
-        test_doc_path = self.fixture_path / "unicode-test.json"
-
-        with open(test_doc_path, "r") as fp:
-            res = json.load(fp)
-
-        doc = Document.from_json(res)
-        doc.annotate_images([Image.new("RGB", (596, 842))])
-
-        predictor_with_rows = IVILATokenClassificationPredictor.from_pretrained(
-            "allenai/ivila-row-layoutlm-finetuned-s2vl-v2"
-        )
-        predictor_with_rows.predict(doc, subpage_per_run=2)

tests/test_recipes/test_core_recipe.py

@@ -5,6 +5,7 @@
 """
 
 import os
+import pathlib
 import unittest
 
 from papermage.magelib import Document, Entity, Image
@@ -40,88 +41,12 @@ def pformat(dictionary, function):
 
 class TestCoreRecipe(unittest.TestCase):
     def setUp(self):
-        self.pdfpath = os.path.join(os.path.dirname(__file__), "../fixtures/1903.10676.pdf")
+        self.fixture_path = pathlib.Path(__file__).parent / "tests/fixtures"
         self.recipe = CoreRecipe()
-        self.doc = self.recipe.from_path(pdfpath=self.pdfpath)
 
-    def test_correct_output(self):
-        self.assertEqual(self.doc.symbols[:1000], FIRST_1000_SYMBOLS)
-        self.assertDictEqual(self.doc.pages[0].to_json(), PAGE_JSON)
-        self.assertEqual(self.doc.images[0].to_base64(), BASE64_PAGE_IMAGE)
-        self.assertListEqual(
-            [round_all_floats(t.to_json()) for t in self.doc.tokens[:10]],
-            round_all_floats(FIRST_10_TOKENS_JSON),
-        )
-        self.assertListEqual(
-            [round_all_floats(r.to_json()) for r in self.doc.rows[:5]],
-            round_all_floats(FIRST_5_ROWS_JSON),
-        )
-
-        self.assertListEqual(
-            [round_all_floats(b.to_json()) for b in self.doc.blocks[:3]],
-            round_all_floats(FIRST_3_BLOCKS_JSON),
-        )
-        self.assertListEqual(
-            [round_all_floats(v.to_json()) for v in self.doc.vila_entities[:10]],
-            round_all_floats(FIRST_10_VILA_JSONS),
-        )
-        # self.assertListEqual(
-        #     [round_all_floats(w.to_json()) for w in self.doc.words[895:900]],
-        #     round_all_floats(SEGMENT_OF_WORD_JSONS),
-        # )
-
-    def test_manual_create_using_annotate(self):
-        """
-        This tests whether one can manually reconstruct a Document without using from_json().
-        Annotations on a Document are order-invariant once created, so you can see this since the
-        fields are being annotated in a different order than they were computed.
-        """
-        doc_json = self.doc.to_json(with_images=True)
-
-        doc2 = Document(symbols=doc_json["symbols"], metadata=doc_json["metadata"])
-        assert doc2.symbols == doc_json["symbols"] == self.doc.symbols
-        assert doc2.metadata.to_json() == doc_json["metadata"] == self.doc.metadata.to_json()
-
-        images = [Image.from_base64(img) for img in doc_json["images"]]
-        doc2.annotate_images(images)
-        assert doc2.images[0].to_base64() == doc_json["images"][0] == self.doc.images[0].to_base64()
-
-        rows = [Entity.from_json(entity_json=r) for r in doc_json["entities"]["rows"]]
-        doc2.annotate_entity(field_name="rows", entities=rows)
-        assert (
-            [r.to_json() for r in doc2.rows]
-            == doc_json["entities"]["rows"]
-            == [r.to_json() for r in self.doc.rows]
-        )
-
-        vila_entities = [Entity.from_json(entity_json=v) for v in doc_json["entities"]["vila_entities"]]
-        doc2.annotate_entity(field_name="vila_entities", entities=vila_entities)
-        assert (
-            [v.to_json() for v in doc2.vila_entities]
-            == doc_json["entities"]["vila_entities"]
-            == [v.to_json() for v in self.doc.vila_entities]
-        )
-
-        # words = [Entity.from_json(entity_json=w) for w in doc_json["entities"]["words"]]
-        # doc2.annotate(words=words)
-        # assert (
-        #     [w.to_json() for w in doc2.words]
-        #     == doc_json["entities"]["words"]
-        #     == [w.to_json() for w in self.doc.words]
-        # )
-
-        tokens = [Entity.from_json(entity_json=t) for t in doc_json["entities"]["tokens"]]
-        doc2.annotate_entity(field_name="tokens", entities=tokens)
-        assert (
-            [t.to_json() for t in doc2.tokens]
-            == doc_json["entities"]["tokens"]
-            == [t.to_json() for t in self.doc.tokens]
-        )
-
-        blocks = [Entity.from_json(entity_json=b) for b in doc_json["entities"]["blocks"]]
-        doc2.annotate_entity(field_name="blocks", entities=blocks)
-        assert (
-            [b.to_json() for b in doc2.blocks]
-            == doc_json["entities"]["blocks"]
-            == [b.to_json() for b in self.doc.blocks]
-        )
+    def test_stability(self):
+        doc = self.recipe.from_path(pdfpath=self.fixture_path / "1903.10676.pdf")
+        doc = self.recipe.from_path(pdfpath=self.fixture_path / "2304.02623v1.pdf")
+        doc = self.recipe.from_path(pdfpath=self.fixture_path / "2020.acl-main.447.pdf")
+        doc = self.recipe.from_path(pdfpath=self.fixture_path / "4be952924cd565488b4a239dc6549095029ee578.pdf")
+        doc = self.recipe.from_path(pdfpath=self.fixture_path / "2023.eacl-main.121.pdf")