Rename dataset_dict.py to dataset.py

dataset.py

@@ -0,0 +1,76 @@
+import datasets
+
+_CITATION = """\
+
+"""
+
+_DESCRIPTION = """\
+VietDrift is a dataset for detecting semantic shift in Vietnamese paraphrases. \
+It has two phases: Phase 1 is binary classification (shift/no-shift), \
+and Phase 2 is multi-label classification of 9 fine-grained shift types.
+"""
+
+_HOMEPAGE = "https://huggingface.co/datasets/ngwgsang/vietdrift"
+
+_LICENSE = "MIT"
+
+_PHASE_CONFIGS = {
+    "phase_1": {
+        "features": datasets.Features({
+            "id": datasets.Value("string"),
+            "sentence1": datasets.Value("string"),
+            "sentence2": datasets.Value("string"),
+            "topic": datasets.Value("string"),
+            "coarse_label": datasets.Value("bool"),
+        }),
+    },
+    "phase_2": {
+        "features": datasets.Features({
+            "id": datasets.Value("string"),
+            "sentence1": datasets.Value("string"),
+            "sentence2": datasets.Value("string"),
+            "topic": datasets.Value("string"),
+            "fine_labels": datasets.Sequence(datasets.Value("string")),
+            "agent_ratio": datasets.Value("float32"),
+            "theme_ratio": datasets.Value("float32"),
+            "action_ratio": datasets.Value("float32"),
+            "recipient_ratio": datasets.Value("float32"),
+            "time_ratio": datasets.Value("float32"),
+            "location_ratio": datasets.Value("float32"),
+            "cause_ratio": datasets.Value("float32"),
+            "instrument_ratio": datasets.Value("float32"),
+            "modality_ratio": datasets.Value("float32"),
+        }),
+    }
+}
+
+class VietDriftConfig(datasets.BuilderConfig):
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+
+class VietDrift(datasets.GeneratorBasedBuilder):
+    BUILDER_CONFIGS = [
+        VietDriftConfig(name="phase_1", version=datasets.Version("1.0.0"), description="Coarse label binary classification"),
+        VietDriftConfig(name="phase_2", version=datasets.Version("1.0.0"), description="Fine-grained multi-label classification"),
+    ]
+
+    DEFAULT_CONFIG_NAME = "phase_1"
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=_PHASE_CONFIGS[self.config.name]["features"],
+            supervised_keys=None,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        data_dir = f"{self.config.name}"
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={"filepath": dl_manager.download_and_extract(f"{data_dir}/train.csv")},
+            ),
+            datasets.SplitGenerator(

dataset_dict.py

@@ -1,9 +0,0 @@
-# vietdrift/dataset_dict.py
-
-from datasets import DatasetDict, load_dataset
-
-def get_dataset_dict(config_name):
-    return load_dataset("csv", data_files={
-        "train": f"{config_name}/train.csv",
-        "test": f"{config_name}/test.csv"
-    })