filtering_posts.py

from __future__ import annotations

import json
import math
import re
from pathlib import Path

import pandas as pd
from rich.console import Console
from rich.progress import BarColumn, Progress, TaskProgressColumn, TextColumn, TimeElapsedColumn
from rich.table import Table


INPUT_DIR = Path("data/posts")
OUTPUT_DIR = Path("data/filtered")
BAN_TAGS_FILE = Path("ban_tags.json")
OUTPUT_FILE_COUNT = 10
ALLOWED_FILE_EXTENSIONS = {"jpg", "jpeg", "png", "webp"}
MIN_IMAGE_WIDTH = 512
MIN_IMAGE_HEIGHT = 512
MIN_TAG_COUNT = 5

# フィルタリングに必要なカラム
NEED_COLUMNS = [
    "id",
    "created_at",
    "md5",
    "score",
    "rating",
    "image_width",
    "image_height",
    "file_ext",
    "tag_count",
    "is_deleted",
    "is_banned",
    "is_pending",
    "is_flagged",
    "tag_string_general",
    "tag_string_character",
    "tag_string_copyright",
    "tag_string_artist",
    "file_url",
]

TAG_COLUMNS = [
    "tag_string_general",
    "tag_string_character",
    "tag_string_copyright",
    "tag_string_artist",
]

# 出力時に必要なカラム
OUTPUT_COLUMNS = [
    "id",
    "created_at",
    "score",
    "rating",
    "tag_count",
    "tag_string_general",
    "tag_string_character",
    "tag_string_copyright",
    "tag_string_artist",
    "file_url",
]

INTERNAL_COLUMNS = ["md5", *OUTPUT_COLUMNS]

console = Console()


def list_post_files() -> list[Path]:
    return sorted(INPUT_DIR.glob("*.parquet"))


def load_ban_tags() -> set[str]:
    with BAN_TAGS_FILE.open(mode="r", encoding="utf-8") as file:
        grouped_tags: dict[str, list[str]] = json.load(file)

    return {
        tag
        for tags in grouped_tags.values()
        for tag in tags
        if tag
    }


def build_tag_pattern(ban_tags: set[str]) -> re.Pattern[str] | None:
    if not ban_tags:
        return None

    escaped_tags = [re.escape(tag) for tag in sorted(ban_tags, key=len, reverse=True)]
    return re.compile(r"(?<!\S)(?:" + "|".join(escaped_tags) + r")(?!\S)")


def load_post_file(path: Path) -> pd.DataFrame:
    return pd.read_parquet(path, columns=NEED_COLUMNS)


def build_filter_mask(data: pd.DataFrame, ban_tag_pattern: re.Pattern[str] | None) -> pd.Series:
    mask = (
        data["score"].ge(0)
        & data["is_deleted"].eq(False)
        & data["is_banned"].eq(False)
        & data["is_pending"].eq(False)
        & data["is_flagged"].eq(False)
        & data["file_ext"].fillna("").str.lower().isin(ALLOWED_FILE_EXTENSIONS)
        & data["image_width"].ge(MIN_IMAGE_WIDTH)
        & data["image_height"].ge(MIN_IMAGE_HEIGHT)
        & data["tag_count"].ge(MIN_TAG_COUNT)
        & data["file_url"].notna()
        & data["file_url"].str.len().gt(0)
    )

    if ban_tag_pattern is None:
        return mask

    has_banned_tag = pd.Series(False, index=data.index)
    for column in TAG_COLUMNS:
        has_banned_tag |= data[column].fillna("").str.contains(ban_tag_pattern, regex=True)

    return mask & ~has_banned_tag


def filter_posts(data: pd.DataFrame, ban_tag_pattern: re.Pattern[str] | None) -> pd.DataFrame:
    mask = build_filter_mask(data, ban_tag_pattern)
    return data.loc[mask, INTERNAL_COLUMNS].reset_index(drop=True)


def filter_post_file(path: Path, ban_tag_pattern: re.Pattern[str] | None) -> tuple[pd.DataFrame, int, int]:
    data = load_post_file(path)
    filtered = filter_posts(data, ban_tag_pattern)
    return filtered, len(data), len(filtered)


def split_ranges(total_rows: int, part_count: int) -> list[tuple[int, int]]:
    if total_rows == 0:
        return [(0, 0) for _ in range(part_count)]

    base_size = total_rows // part_count
    remainder = total_rows % part_count
    ranges: list[tuple[int, int]] = []
    start = 0

    for index in range(part_count):
        size = base_size + (1 if index < remainder else 0)
        end = start + size
        ranges.append((start, end))
        start = end

    return ranges


def save_split_files(data: pd.DataFrame, part_count: int) -> list[tuple[Path, int]]:
    digits = max(2, int(math.log10(part_count - 1)) + 1 if part_count > 1 else 1)
    outputs: list[tuple[Path, int]] = []

    for index, (start, end) in enumerate(split_ranges(len(data), part_count)):
        output_path = OUTPUT_DIR / f"filtered_posts_{index:0{digits}d}.parquet"
        part = data.iloc[start:end].loc[:, OUTPUT_COLUMNS].reset_index(drop=True)
        part.to_parquet(output_path, index=False)
        outputs.append((output_path, len(part)))

    return outputs


def drop_duplicate_posts(data: pd.DataFrame) -> tuple[pd.DataFrame, dict[str, int]]:
    id_deduplicated = data.drop_duplicates(subset="id", keep="first")
    id_duplicate_count = len(data) - len(id_deduplicated)

    md5_values = id_deduplicated["md5"].fillna("")
    has_md5 = md5_values.str.len().gt(0)
    duplicated_md5 = has_md5 & id_deduplicated["md5"].duplicated()
    md5_duplicate_count = int(duplicated_md5.sum())
    deduplicated = id_deduplicated.loc[~duplicated_md5].reset_index(drop=True)

    return deduplicated, {
        "id": id_duplicate_count,
        "md5": md5_duplicate_count,
        "total": len(data) - len(deduplicated),
    }


def print_filter_summary(results: list[tuple[Path, int, int]]) -> None:
    table = Table(title="Filtering Summary")
    table.add_column("file")
    table.add_column("input", justify="right")
    table.add_column("output", justify="right")
    table.add_column("removed", justify="right")

    total_input = 0
    total_output = 0
    for input_path, input_count, output_count in results:
        total_input += input_count
        total_output += output_count
        table.add_row(
            input_path.name,
            f"{input_count:,}",
            f"{output_count:,}",
            f"{input_count - output_count:,}",
        )

    table.add_section()
    table.add_row(
        "total",
        f"{total_input:,}",
        f"{total_output:,}",
        f"{total_input - total_output:,}",
    )
    console.print(table)


def print_output_summary(outputs: list[tuple[Path, int]]) -> None:
    table = Table(title="Output Summary")
    table.add_column("file")
    table.add_column("rows", justify="right")

    for output_path, row_count in outputs:
        table.add_row(output_path.name, f"{row_count:,}")

    console.print(table)


def main() -> None:
    post_files = list_post_files()
    if not post_files:
        raise FileNotFoundError(f"No parquet files found in {INPUT_DIR}")

    OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
    ban_tags = load_ban_tags()
    ban_tag_pattern = build_tag_pattern(ban_tags)
    filtered_frames: list[pd.DataFrame] = []
    results: list[tuple[Path, int, int]] = []

    console.print(f"[bold]Input:[/] {INPUT_DIR}")
    console.print(f"[bold]Output:[/] {OUTPUT_DIR}")
    console.print(f"[bold]Ban tags:[/] {len(ban_tags)}")
    console.print(
        "[bold]Filters:[/] "
        f"ext={sorted(ALLOWED_FILE_EXTENSIONS)}, "
        f"min_size={MIN_IMAGE_WIDTH}x{MIN_IMAGE_HEIGHT}, "
        f"min_tag_count={MIN_TAG_COUNT}"
    )

    with Progress(
        TextColumn("[progress.description]{task.description}"),
        BarColumn(),
        TaskProgressColumn(),
        TimeElapsedColumn(),
        console=console,
    ) as progress:
        task = progress.add_task("filtering parquet files", total=len(post_files))

        for path in post_files:
            progress.update(task, description=path.name)
            filtered, input_count, output_count = filter_post_file(path, ban_tag_pattern)
            filtered_frames.append(filtered)
            results.append((path, input_count, output_count))
            progress.advance(task)

    print_filter_summary(results)

    console.print("[bold]Concatenating filtered posts...[/]")
    filtered_posts = pd.concat(filtered_frames, ignore_index=True)
    filtered_posts, duplicate_counts = drop_duplicate_posts(filtered_posts)
    console.print(
        "[bold]Duplicates removed:[/] "
        f"id={duplicate_counts['id']:,}, "
        f"md5={duplicate_counts['md5']:,}, "
        f"total={duplicate_counts['total']:,}"
    )

    console.print(f"[bold]Saving:[/] {OUTPUT_FILE_COUNT} parquet files")
    outputs = save_split_files(filtered_posts, OUTPUT_FILE_COUNT)
    print_output_summary(outputs)


if __name__ == "__main__":
    main()