dataset: add initial version of loader

HisGermaNER.py

@@ -0,0 +1,113 @@
+import datasets
+
+# Coming soon!
+_CITATION = ""
+
+_DESCRIPTION = """\
+HisGermaNER is another NER dataset from historical German newspapers.
+
+In the first release of our dataset, 11 newspapers from 1710 to 1840 from the Austrian National Library (ONB) are selected, resulting in 100 pages.
+"""
+
+
+class HisGermaNERConfig(datasets.BuilderConfig):
+    """BuilderConfig for HisGermaNER"""
+
+    def __init__(self, data_url, **kwargs):
+        super(HisGermaNERConfig, self).__init__(**kwargs)
+        self.data_url = data_url
+
+
+class HisGermaNER(datasets.GeneratorBasedBuilder):
+    BUILDER_CONFIGS = [
+        HisGermaNERConfig(
+            name="HisGermaNER",
+            version=datasets.Version("0.0.1"),
+            description="HisGermaNER Dataset",
+            data_url="https://huggingface.co/datasets/stefan-it/HisGermaNER/resolve/main/splits/HisGermaNER_v0_",
+        )
+    ]
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features(
+                {
+                    "id": datasets.Value("string"),
+                    "tokens": datasets.Sequence(datasets.Value("string")),
+                    "ner_tags": datasets.Sequence(
+                        datasets.features.ClassLabel(
+                            names=[
+                                "O",
+                                "B-PER",
+                                "I-PER",
+                                "B-ORG",
+                                "I-ORG",
+                                "B-LOC",
+                                "I-LOC",
+                            ]
+                        )
+                    )
+                }
+            ),
+            supervised_keys=None,
+            homepage="https://huggingface.co/datasets/stefan-it/HisGermaNER",
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        """Returns generator for dataset splits."""
+        download_urls = {
+            split: self.config.data_url + split + ".tsv" for split in ["train", "dev", "test"]
+        }
+
+        downloaded_files = dl_manager.download_and_extract(download_urls)
+
+        splits = [
+            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["train"]}),
+            datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files["dev"]}),
+            datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": downloaded_files["test"]})
+        ]
+
+        return splits
+
+    def _generate_examples(self, filepath):
+        with open(filepath, "rt", encoding="utf-8") as f_p:
+            current_tokens = []
+            current_tags = []
+
+            sentence_counter = 0
+
+            for line in f_p:
+                line = line.strip()
+                if not line:
+                    if len(current_tokens) > 0:
+                        sentence = (
+                            sentence_counter, {
+                                "id": str(sentence_counter),
+                                "tokens": current_tokens,
+                                "ner_tags": current_tags,
+                            }
+                        )
+                        sentence_counter += 1
+                        current_tokens = []
+                        current_tags = []
+                        yield sentence
+                    continue
+
+                if line.startswith("TOKEN"):
+                    continue
+
+                if line.startswith("# "):
+                    continue
+
+                token, tag, misc = line.split("\t")
+                current_tokens.append(token)
+                current_tags.append(tag)
+
+            if len(current_tokens) > 0:
+                yield sentence_counter, {
+                    "id": str(sentence_counter),
+                    "tokens": current_tokens,
+                    "ner_tags": current_tags,
+                }