dataset.py

# import os
# import csv
# from dataset import GeneratorBasedBuilder, DatasetInfo, BuilderConfig, SplitGenerator, Split, Features, Value

# _LANGUAGES = ["AR", "CS", "EN", "ZH", "HU", "RU", "SR", "KO", "TI", "VI"]
# _TASKS_WITH_SETTINGS = ["MDUR", "MIQA", "MMJB"]
# _SETTINGS = ["traditional", "vision"]

# class PM4BenchConfig(BuilderConfig):
#     def __init__(self, task, setting, language, **kwargs):
#         super().__init__(**kwargs)
#         self.task = task
#         self.setting = setting
#         self.language = language

# class PM4Bench(GeneratorBasedBuilder):
#     BUILDER_CONFIG_CLASS = PM4BenchConfig

#     BUILDER_CONFIGS = []
#     for task in _TASKS_WITH_SETTINGS:
#         for setting in _SETTINGS:
#             for lang in _LANGUAGES:
#                 BUILDER_CONFIGS.append(
#                     PM4BenchConfig(
#                         name=f"{task}_{setting}_{lang}",
#                         task=task,
#                         setting=setting,
#                         language=lang,
#                         version="1.0.0",
#                         description=f"{task} task with {setting} setting in {lang} language"
#                     )
#                 )
#     for lang in _LANGUAGES:
#         BUILDER_CONFIGS.append(
#             PM4BenchConfig(
#                 name=f"MSOCR_vision_{lang}",
#                 task="MSOCR",
#                 setting="vision",
#                 language=lang,
#                 version="1.0.0",
#                 description=f"MSOCR task with vision setting in {lang} language"
#             )
#         )

#     def _info(self):
#         # Use placeholder; actual features will be inferred dynamically at runtime
#         return DatasetInfo(
#             description="PM4Bench: Multilingual benchmark for multiple tasks with vision and traditional settings.",
#             features=Features({}),
#             supervised_keys=None,
#             homepage="",
#             license="",
#         )

#     def _split_generators(self, dl_manager):
#         cfg = self.config
#         if cfg.task == "MSOCR":
#             file_path = os.path.join("MSOCR", f"MSOCR_{cfg.setting}_{cfg.language}.tsv")
#         else:
#             file_path = os.path.join(cfg.task, cfg.setting, f"{cfg.task}_{cfg.setting}_{cfg.language}.tsv")

#         return [
#             SplitGenerator(
#                 name=Split.TRAIN,
#                 gen_kwargs={"filepath": file_path}
#             )
#         ]

#     def _generate_examples(self, filepath):
#         with open(filepath, encoding="utf-8") as f:
#             reader = csv.DictReader(f, delimiter="\t")
#             headers = reader.fieldnames
#             # Skip 'index' column if present
#             if headers and headers[0].lower() == "index":
#                 headers = headers[1:]

#             for idx, row in enumerate(reader):
#                 # Remove 'index' field if exists
#                 example = {key: row[key] for key in headers if key in row}
#                 yield idx, example
import os
import csv
from datasets import GeneratorBasedBuilder, DatasetInfo, BuilderConfig, SplitGenerator, Split, Features, Value

class PM4Bench(GeneratorBasedBuilder):

    def _generate_examples(self, filepath):
        with open(filepath, encoding="utf-8") as f:
            reader = csv.DictReader(f, delimiter="\t")
            headers = reader.fieldnames
            # Skip 'index' column if present
            if headers and headers[0].lower() == "index":
                headers = headers[1:]

            for idx, row in enumerate(reader):
                if idx >= 100:  # Limit preview to first 100 rows
                    break
                # Remove 'index' field if exists
                example = {key: row[key] for key in headers if key in row}
                yield idx, example