economy-watchers-survey-evaluation.py

import datasets
from datasets import load_dataset

# TODO: Add BibTeX citation
# Find for instance the citation on arxiv or on the dataset repo/website
_CITATION = """\
TBA
"""

_DESCRIPTION = """\
Dataset from the Economy Watchers Survey for use as evaluation tasks
"""

_HOMEPAGE = "https://github.com/retarfi/economy-watchers-survey"
_LICENSE = "CC-BY 4.0"
VERSION = datasets.Version("0.1.2")
EWS_REPO_NAME = "retarfi/economy-watchers-survey"

LABEL_REASON_OTHER: str = "それ以外"
DISABLED_BARS: bool = datasets.are_progress_bars_disabled()


class EconomyWatchersSurveyConfig(datasets.BuilderConfig):
    def __init__(self, **kwargs):
        self.ews_version = kwargs.pop("ews_version", None)
        super(EconomyWatchersSurveyConfig, self).__init__(**kwargs)


class EconomyWatchersSurveyDataset(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIG_CLASS = EconomyWatchersSurveyConfig

    BUILDER_CONFIGS = [
        EconomyWatchersSurveyConfig(
            name="sentiment", version=VERSION, description="Sentiment analysis"
        ),
        EconomyWatchersSurveyConfig(
            name="domain", version=VERSION, description="Domain classification"
        ),
        EconomyWatchersSurveyConfig(
            name="reason",
            version=VERSION,
            description="Classification of reason to decision of sentiment",
        ),
    ]

    def _info(self):
        feat_label: datasets.ClassLabel
        if self.config.name == "sentiment":
            feat_label = datasets.ClassLabel(
                num_classes=5, names=["×", "▲", "□", "○", "◎"]
            )
        elif self.config.name == "domain":
            feat_label = datasets.ClassLabel(
                num_classes=3, names=["家計動向", "企業動向", "雇用"]
            )
        elif self.config.name == "reason":
            feat_label = datasets.ClassLabel(
                num_classes=12,
                names=[
                    "来客数の動き",
                    "販売量の動き",
                    "お客様の様子",
                    "受注量や販売量の動き",
                    "単価の動き",
                    "取引先の様子",
                    "求人数の動き",
                    "競争相手の様子",
                    "受注価格や販売価格の動き",
                    "周辺企業の様子",
                    "求職者数の動き",
                    LABEL_REASON_OTHER,
                ],
            )
        else:
            raise NotImplementedError(self.config.name)
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "text": datasets.Value("string"),
                    "label": feat_label,
                    "id": datasets.Value("string"),
                }
            ),
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(
        self, dl_manager: datasets.DownloadManager
    ) -> list[datasets.SplitGenerator]:
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN, gen_kwargs={"split": "train"}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION, gen_kwargs={"split": "validation"}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST, gen_kwargs={"split": "test"}
            ),
        ]

    def _generate_examples(self, split: str):
        batch_size = 1000
        
        ds_info_current = load_dataset(
            EWS_REPO_NAME, "current", revision=self.config.ews_version, split=split
        )
        total_current = len(ds_info_current)
        
        if self.config.name == "reason":
            for start_idx in range(0, total_current, batch_size):
                end_idx = min(start_idx + batch_size, total_current)
                ds_current_batch = load_dataset(
                    EWS_REPO_NAME, "current", revision=self.config.ews_version, 
                    split=f"{split}[{start_idx}:{end_idx}]"
                )
                
                if not DISABLED_BARS:
                    datasets.disable_progress_bars()
                ds_current_batch = ds_current_batch.filter(lambda example: example["判断の理由"] is not None)
                if not DISABLED_BARS:
                    datasets.enable_progress_bars()
                for example in ds_current_batch:
                    str_label = example["判断の理由"]
                    if str_label not in self.info.features["label"].names:
                        str_label = LABEL_REASON_OTHER
                    
                    yield example["id"], {
                        "id": example["id"],
                        "text": example["追加説明及び具体的状況の説明"],
                        "label": self.info.features["label"].str2int(str_label),
                    }
        else:
            ds_info_future = load_dataset(
                EWS_REPO_NAME, "future", revision=self.config.ews_version, split=split
            )
            total_future = len(ds_info_future)
            for start_idx in range(0, total_current, batch_size):
                end_idx = min(start_idx + batch_size, total_current)
                ds_current_batch = load_dataset(
                    EWS_REPO_NAME, "current", revision=self.config.ews_version, 
                    split=f"{split}[{start_idx}:{end_idx}]"
                )
                
                ds_current_no_reason = ds_current_batch.remove_columns("判断の理由")
                for example in ds_current_no_reason:
                    str_label: str
                    if self.config.name == "sentiment":
                        str_label = example["景気の現状判断"]
                    elif self.config.name == "domain":
                        str_label = example["関連"]
                    
                    yield example["id"], {
                        "id": example["id"],
                        "text": example["追加説明及び具体的状況の説明"],
                        "label": self.info.features["label"].str2int(str_label),
                    }
            
            for start_idx in range(0, total_future, batch_size):
                end_idx = min(start_idx + batch_size, total_future)
                
                ds_future_batch = load_dataset(
                    EWS_REPO_NAME, "future", revision=self.config.ews_version, 
                    split=f"{split}[{start_idx}:{end_idx}]"
                )
                ds_future_renamed = ds_future_batch.rename_columns(
                    {"景気の先行き判断": "景気の現状判断", "景気の先行きに対する判断理由": "追加説明及び具体的状況の説明"}
                )
                for example in ds_future_renamed:
                    str_label: str
                    if self.config.name == "sentiment":
                        str_label = example["景気の現状判断"]
                    elif self.config.name == "domain":
                        str_label = example["関連"]
                    
                    yield example["id"], {
                        "id": example["id"],
                        "text": example["追加説明及び具体的状況の説明"],
                        "label": self.info.features["label"].str2int(str_label),
                    }