hal-summarization.py

import json
import os
import datasets
from tqdm import tqdm


_ARTICLE_ID = "article_id"
_ARTICLE_WORDS = "article_words"
_ARTICLE_BBOXES = "article_bboxes"
_ARTICLE_NORM_BBOXES = "article_norm_bboxes"
_ABSTRACT = "abstract"
_ARTICLE_PDF_URL = "article_pdf_url"

def normalize_bbox(bbox, size):
    return [
        int(1000 * bbox[0] / size[0]),
        int(1000 * bbox[1] / size[1]),
        int(1000 * bbox[2] / size[0]),
        int(1000 * bbox[3] / size[1]),
    ]


class HALSummarizationConfig(datasets.BuilderConfig):
    """BuilderConfig for HALSummarization."""
    def __init__(self, **kwargs):
        """BuilderConfig for ArxivSummarization.
        Args:
          **kwargs: keyword arguments forwarded to super.
        """
        super(HALSummarizationConfig, self).__init__(**kwargs)
        

class HALSummarizationDataset(datasets.GeneratorBasedBuilder):
    """HALSummarization Dataset."""

    _TRAIN_ARCHIVE = "train.zip"
    _VAL_ARCHIVE = "val.zip"
    _TEST_ARCHIVE = "test.zip"
    _TRAIN_ABSTRACTS = "train.txt"
    _VAL_ABSTRACTS = "validation.txt"
    _TEST_ABSTRACTS = "test.txt"
    
    BUILDER_CONFIGS = [
        HALSummarizationConfig(
            name="hal",
            version=datasets.Version("1.0.0"),
            description="HAL dataset for summarization",
        ),
    ]

    
    def _info(self):
        # Should return a datasets.DatasetInfo object
        return datasets.DatasetInfo(
            features=datasets.Features(
                {
                    _ARTICLE_ID: datasets.Value("string"),
                    _ARTICLE_WORDS: datasets.Sequence(datasets.Value("string")),
                    _ARTICLE_BBOXES: datasets.Sequence(datasets.Sequence(datasets.Value("int64"))),
                    _ARTICLE_NORM_BBOXES: datasets.Sequence(datasets.Sequence(datasets.Value("int64"))),
                    _ABSTRACT: datasets.Value("string"),
                    _ARTICLE_PDF_URL: datasets.Value("string"),
                }
            ),
            supervised_keys=None,
        )

    def _split_generators(self, dl_manager):

        train_dir = os.path.join(dl_manager.download_and_extract(self._TRAIN_ARCHIVE), "train")
        val_dir = os.path.join(dl_manager.download_and_extract(self._VAL_ARCHIVE), "val")
        test_dir = os.path.join(dl_manager.download_and_extract(self._TEST_ARCHIVE), "test")

        train_abstracts = dl_manager.download_and_extract(self._TRAIN_ABSTRACTS)
        val_abstracts = dl_manager.download_and_extract(self._VAL_ABSTRACTS)
        test_abstracts = dl_manager.download_and_extract(self._TEST_ABSTRACTS)
        
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN, 
                gen_kwargs={"data_path": train_dir, "abstract_path": train_abstracts}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION, 
                gen_kwargs={"data_path": val_dir, "abstract_path": val_abstracts}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST, 
                gen_kwargs={"data_path": test_dir, "abstract_path": test_abstracts}
            ),
        ]

    def _generate_examples(self, data_path, abstract_path):
        """Generate HALSummarization examples."""
        filenames = sorted(os.listdir(data_path))

        guid = 0
        with open(abstract_path, 'r') as abstract_file:
            for line in tqdm(abstract_file, total=len(filenames), desc=f"Reading files in {data_path}"):
                guid += 1
                item = json.loads(line)
                fname = item["id"] + ".txt"
                filepath = os.path.join(data_path, fname)
                
                words  = []
                bboxes = []
                norm_bboxes = []

                with open(filepath, encoding="utf-8") as f:
                    for line in f:
                        splits = line.split("\t")
                        word = splits[0]
                        bbox = splits[1:5]
                        bbox = [int(b) for b in bbox]
                        page_width, page_height = int(splits[5]), int(splits[6])
                        norm_bbox = normalize_bbox(bbox, (page_width, page_height))

                        words.append(word)
                        bboxes.append(bbox)
                        norm_bboxes.append(norm_bbox)

                assert len(words) == len(bboxes)
                assert len(bboxes) == len(norm_bboxes)

                yield guid, {
                        _ARTICLE_ID: item["id"],
                        _ARTICLE_WORDS: words, 
                        _ARTICLE_BBOXES: bboxes, 
                        _ARTICLE_NORM_BBOXES: norm_bboxes,
                        _ABSTRACT: item["abstract"],
                        _ARTICLE_PDF_URL: item["pdf_url"],
                    }