Add dataset files

README.md

@@ -0,0 +1,3 @@
+This is a in-house development version of KLUE Topic Classification benchmark, as the test split is not released by the KLUE team.
+
+We randomly split the original validation set (9,107 instances) into in-house validation set (5,107 instances) and the in-house test set (4,000 instances).

data/label.txt

@@ -0,0 +1,7 @@
+정치
+세계
+IT과학
+스포츠
+사회
+경제
+생활문화

klue-tc-dev-tsv.py

@@ -0,0 +1,55 @@
+from __future__ import absolute_import, division, print_function
+import datasets
+_URL = "data/"
+_URLs = {
+    "train": _URL + "train.tsv",
+    "valid": _URL + "valid.tsv",
+    "test": _URL + "test.tsv",
+}
+
+
+class KlueTC(datasets.GeneratorBasedBuilder):
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description="KLUE Topic Classification (Dev Split)",
+            features=datasets.Features(
+                {
+                    "text": datasets.Value("string"),
+                    "label": datasets.features.ClassLabel(names=['정치', '세계', 'IT과학', '스포츠', '사회', '경제', '생활문화']),
+                }
+            ),
+            supervised_keys=None,
+            license="",
+            homepage="",
+            citation="",
+        )
+        
+    def _split_generators(self, dl_manager):
+        downloaded_files = dl_manager.download_and_extract(_URLs)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "filepath": downloaded_files["train"],
+                }
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                gen_kwargs={
+                    "filepath": downloaded_files["valid"],
+                }
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST,
+                gen_kwargs={
+                    "filepath": downloaded_files["test"],
+                }
+            ),
+        ]
+        
+    def _generate_examples(self, filepath):
+        with open(filepath, "r", encoding='UTF-8') as f:
+            for idx, line in enumerate(f):
+                text, label = line.split("\t")
+                yield idx, {"text": text.strip(), "label": label.strip()}