Fix the entailment script.

scifact_entailment.py

@@ -2,9 +2,8 @@
 using evidence from the cited abstracts. Formatted as a paragraph-level entailment task."""
 
 
-import json
-
 import datasets
+import json
 
 
 _CITATION = """\
@@ -20,6 +19,12 @@ _DESCRIPTION = """\
 SciFact, a dataset of 1.4K expert-written scientific claims paired with evidence-containing abstracts, and annotated with labels and rationales.
 """
 
+_URL = "https://scifact.s3-us-west-2.amazonaws.com/release/latest/data.tar.gz"
+
+
+def flatten(xss):
+    return [x for xs in xss for x in xs]
+
 
 class ScifactEntailmentConfig(datasets.BuilderConfig):
     """BuilderConfig for Scifact"""
@@ -43,17 +48,15 @@ class ScifactEntailment(datasets.GeneratorBasedBuilder):
 
     def _info(self):
         # TODO(scifact): Specifies the datasets.DatasetInfo object
+
         features = {
-            "id": datasets.Value("int32"),  # An integer claim ID.
-            "claim": datasets.Value("string"),  # The text of the claim.
-            "evidence_doc_id": datasets.Value("string"),
-            "evidence_label": datasets.Value("string"),  # Label for the rationale.
-            "evidence_sentences": datasets.features.Sequence(
-                datasets.Value("int32")
-            ),  # Rationale sentences.
-            "cited_doc_ids": datasets.features.Sequence(
-                datasets.Value("int32")
-            ),  # The claim's "cited documents".
+            "claim_id": datasets.Value("int32"),
+            "claim": datasets.Value("string"),
+            "abstract_id": datasets.Value("int32"),
+            "title": datasets.Value("string"),
+            "abstract": datasets.features.Sequence(datasets.Value("string")),
+            "verdict": datasets.Value("string"),
+            "evidence": datasets.features.Sequence(datasets.Value("int32")),
         }
 
         return datasets.DatasetInfo(
@@ -73,74 +76,82 @@ class ScifactEntailment(datasets.GeneratorBasedBuilder):
             citation=_CITATION,
         )
 
+    @staticmethod
+    def _read_tar_file(f):
+        res = []
+        for row in f:
+            this_row = json.loads(row.decode("utf-8"))
+            res.append(this_row)
+
+        return res
+
     def _split_generators(self, dl_manager):
         """Returns SplitGenerators."""
         # TODO(scifact): Downloads the data and defines the splits
         # dl_manager is a datasets.download.DownloadManager that can be used to
         # download and extract URLs
+        archive = dl_manager.download(_URL)
+        for path, f in dl_manager.iter_archive(archive):
+            if path == "data/corpus.jsonl":
+                corpus = self._read_tar_file(f)
+                corpus = {x["doc_id"]: x for x in corpus}
+            elif path == "data/claims_train.jsonl":
+                claims_train = self._read_tar_file(f)
+            elif path == "data/claims_dev.jsonl":
+                claims_validation = self._read_tar_file(f)
+
         return [
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
                 # These kwargs will be passed to _generate_examples
                 gen_kwargs={
+                    "claims": claims_train,
+                    "corpus": corpus,
                     "split": "train",
                 },
             ),
-            datasets.SplitGenerator(
-                name=datasets.Split.TEST,
-                # These kwargs will be passed to _generate_examples
-                gen_kwargs={
-                    "split": "test",
-                },
-            ),
             datasets.SplitGenerator(
                 name=datasets.Split.VALIDATION,
                 # These kwargs will be passed to _generate_examples
                 gen_kwargs={
-                    "split": "dev",
+                    "claims": claims_validation,
+                    "corpus": corpus,
+                    "split": "validation",
                 },
             ),
         ]
 
-    def _generate_examples(self, split):
+    def _generate_examples(self, claims, corpus, split):
         """Yields examples."""
-        # TODO(scifact): Yields (key, example) tuples from the dataset
-
-        # Load corpus and convert to dict.
-        corpus = datasets.load_dataset("bigbio/scifact", "scifact_corpus_source", split="train")
-        corpus = {x["doc_id"]: x for x in corpus}
-
-        # Load claims.
-        claims = datasets.load_dataset("bigbio/scifact", "scifact_claims_source", split=split)
-
-        for id_, claim in enumerate(claims):
-            evidence = {x["doc_id"]: x for x in claim["evidences"]}
+        # Loop over claims and put evidence together with claim.
+        id_ = -1  # Will increment to 0 on first iteration.
+        for claim in claims:
+            evidence = {int(k): v for k, v in claim["evidence"].items()}
             for cited_doc_id in claim["cited_doc_ids"]:
                 cited_doc = corpus[cited_doc_id]
-                # Format the abstract.
-                sent_ids = [f"[{i}]" for i in range(len(cited_doc["abstract"]))]
-                # Get rid of newlines.
-                sents = [sent.strip() for sent in cited_doc["abstract"]]
-                zipped = zip(sent_ids, sents)
-                cited_abstract = " ".join(
-                    [f"{entry[0]} {entry[1]}" for entry in zipped]
-                )
+                abstract_sents = [sent.strip() for sent in cited_doc["abstract"]]
 
                 if cited_doc_id in evidence:
-                    verdict = evidence[cited_doc_id]["label"]
-                    sents = evidence[cited_doc_id]["sentence_ids"]
+                    this_evidence = evidence[cited_doc_id]
+                    verdict = this_evidence[0][
+                        "label"
+                    ]  # Can take first evidence since all labels are same.
+                    evidence_sents = flatten(
+                        [entry["sentences"] for entry in this_evidence]
+                    )
                 else:
                     verdict = "NEI"
-                    sents = []
+                    evidence_sents = []
 
                 instance = {
-                    "id": claim["id"],
+                    "claim_id": claim["id"],
                     "claim": claim["claim"],
                     "abstract_id": cited_doc_id,
-                    "title":  cited_doc["title"],
-                    "abstract": cited_abstract,
+                    "title": cited_doc["title"],
+                    "abstract": abstract_sents,
                     "verdict": verdict,
-                    "evidence": sents,
+                    "evidence": evidence_sents,
                 }
 
+                id_ += 1
                 yield id_, instance