murakami.py

"""
Parse all paragraphs from all *.fb2 files in the input directory, create a Huggingface Dataset and push it to the Hub as `vldsavelyev/murakami`.
"""


import os
from pathlib import Path
from lxml import etree
import datasets
from datasets import Dataset
from huggingface_hub import create_repo

datasets.logging.set_verbosity_info()


_DESCRIPTION = """\
Russian translations of Murakami novels, to fine-tune a generative language model. Source is FB2 files from http://flibusta.is/a/8570.
"""


class Builder(datasets.GeneratorBasedBuilder):
    """Murakami novels, translated to Russian."""

    VERSION = datasets.Version("1.1.0")

    # This is an example of a dataset with multiple configurations.
    # If you don't want/need to define several sub-sets in your dataset,
    # just remove the BUILDER_CONFIG_CLASS and the BUILDER_CONFIGS attributes.

    # If you need to make complex sub-parts in the datasets with configurable options
    # You can create your own builder configuration class to store attribute, inheriting from datasets.BuilderConfig
    # BUILDER_CONFIG_CLASS = MyBuilderConfig

    # You will be able to load one or the other configurations in the following list with
    # data = datasets.load_dataset('my_dataset', 'first_domain')
    # data = datasets.load_dataset('my_dataset', 'second_domain')

    def _info(self):
        # This method specifies the datasets.DatasetInfo object which contains informations and typings for the dataset
        return datasets.DatasetInfo(
            # This is the description that will appear on the datasets page.
            description=_DESCRIPTION,
            # This defines the different columns of the dataset and their types
            features=datasets.Features({"text": datasets.Value("string")}),
        )

    # Number of initial <p> element to take from each fb2, by number. This allows to skip
    # intros and other junk in the beginning of an fb2. This is built semi-manually using
    # the `helper_to_find_first_paragraphs` func.
    START_PARAGRAPHS = {
        3: 5,
        6: 27,
        7: 3,
        9: 4,
        10: 3,
        12: 11,
        18: 5,
        20: 3,
        21: 5,
    }

    @staticmethod
    def helper_to_find_first_paragraphs(paragraphs, title, book_number, n=30):
        """
        Helps to eyeball first few paragraphs of a book to skip junk paragraphs
        in the beginning and manually construct the `tart_paragraphs` dict.
        """
        found_paragraphs = []
        skipping = True
        for i, p in enumerate(list(paragraphs)[:n]):
            if p.text is None:
                continue
            if (
                book_number in Builder.START_PARAGRAPHS
                and i >= Builder.START_PARAGRAPHS[book_number]
            ):
                skipping = False
            if skipping and p.text.lower() == title.lower():
                skipping = False
            if not skipping:
                found_paragraphs.append(f"   {i} {p.text}")

        if found_paragraphs:
            print("✅")
            print("\n".join(found_paragraphs))

        else:
            print("❌")
            for i, p in enumerate(list(paragraphs)[:30]):
                print(f"   {i} {p.text}")

    def _split_generators(self, dl_manager):
        # This method is tasked with downloading/extracting the data and defining the splits depending on the configuration
        # If several configurations are possible (listed in BUILDER_CONFIGS), the configuration selected by the user is in self.config.name
        data_dir = "data"
        text_by_name = {}
        fb2s = list(Path(data_dir).glob("*.fb2"))
        if len(fb2s) > 0:
            print(f"Found {len(fb2s)} fb2 files in {data_dir}")
        else:
            raise ValueError(f"No fb2 files found in {data_dir}")

        for bi, path in enumerate(fb2s):
            print(bi, path)

            # Load the FB2 format file
            with path.open("rb") as file:
                fb2_data = file.read()

            # Print structure of the FB2 format file
            # print(etree.tostring(etree.fromstring(fb2_data), pretty_print=True))

            # Parse the FB2 format file using lxml
            root = etree.fromstring(fb2_data)

            # Get the title of the book
            title = root.xpath(
                "//fb:title-info/fb:book-title",
                namespaces={"fb": "http://www.gribuser.ru/xml/fictionbook/2.0"},
            )[0].text
            print(title)

            # Get all book paragraphs
            paragraphs = root.xpath(
                "//fb:p",
                namespaces={"fb": "http://www.gribuser.ru/xml/fictionbook/2.0"},
            )

            # UNCOMMENT THE LINE BELOW TO BUILD `START_PARAGRAPHS`:
            # self.helper_to_find_first_paragraphs(paragraphs, title, bi)

            found_paragraphs = []
            skipping = True
            for pi, p in enumerate(paragraphs):
                if p.text is None:
                    continue
                if (
                    bi in Builder.START_PARAGRAPHS
                    and pi >= Builder.START_PARAGRAPHS[bi]
                ):
                    skipping = False
                if skipping and p.text.lower() == title.lower():
                    skipping = False
                if not skipping:
                    found_paragraphs.append(p)
            print(f"Found {len(found_paragraphs)} paragraphs")

            text_by_name[title] = ""
            for p in found_paragraphs:
                text_by_name[title] += p.text.replace(" ", " ") + "\n"
            text_by_name[title] += "\n"

        print("Novel by size:")
        for title, text in text_by_name.items():
            print(f"  {title}: {len(text):,} characters")

        smallest_title = min(text_by_name, key=lambda k: len(text_by_name[k]))
        print(
            f"Using smallest novel {smallest_title} "
            f"({len(text_by_name[smallest_title]):,} characters) as a test set"
        )

        test_titles = [smallest_title]
        train_titles = [t for t in text_by_name if t not in test_titles]

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                # These kwargs will be passed to _generate_examples
                gen_kwargs={
                    "titles": train_titles,
                    "texts": [text_by_name[t] for t in train_titles],
                    "split": "train",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                # These kwargs will be passed to _generate_examples
                gen_kwargs={
                    "titles": test_titles,
                    "texts": [text_by_name[t] for t in test_titles],
                    "split": "test",
                },
            ),
        ]

    # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
    def _generate_examples(self, titles, texts, split):
        # This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.
        # The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.
        for title, text in zip(titles, texts):
            yield title, {"text": text}