filter.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

"""
Filter Sheetpedia-style XLSX files by worksheet-level quality rules.

Main logic:
1. Iterate over all .xlsx files under input_dir
2. For each worksheet, compute metadata on the effective table region
3. Decide keep/drop for each worksheet
4. Keep the workbook if at least one worksheet is kept
5. Export worksheet-level metadata and file-level summary

Recommended usage:
    python filter_xlsx_dataset.py \
        --input_dir /path/to/xlsx_folder \
        --output_dir /path/to/output_metadata

Dependencies:
    pip install openpyxl pandas
"""

import os
import re
import csv
import math
import argparse
import unicodedata
from collections import Counter
from pathlib import Path

import pandas as pd
from openpyxl import load_workbook


# =========================
# Configuration
# =========================

DEFAULT_MAX_ROWS = 100
DEFAULT_MAX_COLS = 80
DEFAULT_MIN_ROWS = 3
DEFAULT_MIN_COLS = 3
DEFAULT_MIN_FILL_RATIO = 0.15
DEFAULT_MAX_MISSING_NOISE_RATIO = 0.30
DEFAULT_MAX_PLACEHOLDER_RATIO = 0.50
DEFAULT_MIN_HEADER_UNIQUE_RATIO = 0.50
DEFAULT_MAX_NUMERIC_HEADER_RATIO = 0.80

# Missing / noisy markers after normalization
MISSING_MARKERS = {
    "", "na", "n/a", "null", "none", "nan", "#n/a", "-", "--", "?", "unk", "unknown"
}

# Generic low-information placeholder tokens
PLACEHOLDER_TOKENS = {
    "name", "field", "value", "item", "data", "text", "label",
    "column", "row", "header", "col", "attr"
}


# =========================
# Utility functions
# =========================

def normalize_text(value):
    """
    Normalize a cell value into a comparable string.
    """
    if value is None:
        return ""

    # Keep numbers/bools readable
    if isinstance(value, bool):
        s = str(value).lower()
    elif isinstance(value, (int, float)):
        if isinstance(value, float):
            if math.isnan(value):
                return "nan"
            if math.isinf(value):
                return "inf"
        s = str(value)
    else:
        s = str(value)

    # Unicode normalization
    s = unicodedata.normalize("NFKC", s)

    # Remove surrounding whitespace and collapse internal whitespace
    s = s.strip()
    s = re.sub(r"\s+", " ", s)

    return s


def is_empty_cell(value):
    """
    Define whether a cell is empty for effective region detection.
    """
    s = normalize_text(value)
    return s == ""


def is_missing_marker(value):
    """
    Detect common missing-value markers after normalization.
    """
    s = normalize_text(value).lower()
    return s in MISSING_MARKERS


def contains_replacement_char(s):
    return "�" in s


def contains_control_chars(s):
    """
    Keep common whitespace chars; detect suspicious non-printable chars.
    """
    for ch in s:
        cat = unicodedata.category(ch)
        if cat.startswith("C") and ch not in ("\t", "\n", "\r"):
            return True
    return False


def informative_char_ratio(s):
    """
    Ratio of letters/digits/CJK characters among non-space characters.
    """
    s_no_space = re.sub(r"\s+", "", s)
    if not s_no_space:
        return 0.0

    valid_count = 0
    for ch in s_no_space:
        # Latin letters / digits
        if ch.isalnum():
            valid_count += 1
            continue

        # CJK Unified Ideographs
        code = ord(ch)
        if (
            0x4E00 <= code <= 0x9FFF or
            0x3400 <= code <= 0x4DBF or
            0x20000 <= code <= 0x2A6DF
        ):
            valid_count += 1

    return valid_count / len(s_no_space)


def is_symbol_noise(s):
    """
    Detect strings dominated by repeated punctuation/symbols.
    Examples: #####, ****, ..., @@@@
    """
    s = s.strip()
    if len(s) < 3:
        return False

    # all chars same non-alnum symbol
    if len(set(s)) == 1 and not s[0].isalnum():
        return True

    # mostly punctuation/symbols
    non_space = re.sub(r"\s+", "", s)
    if not non_space:
        return False

    punct_or_symbol = sum(
        1 for ch in non_space
        if unicodedata.category(ch)[0] in {"P", "S"}
    )
    return punct_or_symbol / len(non_space) >= 0.8


def is_garbled(value):
    """
    Detect garbled/noisy cell values.
    """
    s = normalize_text(value)
    if s == "":
        return False

    if contains_replacement_char(s):
        return True

    if contains_control_chars(s):
        return True

    if is_symbol_noise(s):
        return True

    # low informative ratio for non-trivial strings
    if len(s) >= 4 and informative_char_ratio(s) < 0.30:
        return True

    return False


def is_numeric_like(s):
    """
    Determine whether a normalized string is numeric-like.
    """
    s = s.strip()
    if s == "":
        return False

    try:
        float(s.replace(",", ""))
        return True
    except Exception:
        return False


def is_placeholder_like(value):
    """
    Detect placeholder-like header content.
    """
    s = normalize_text(value).lower()
    if s == "":
        return False

    # [name], [date], [field], [column1], etc.
    if re.fullmatch(r"\[[^\[\]]+\]", s):
        return True

    # generic tokens
    if s in PLACEHOLDER_TOKENS:
        return True

    # auto-generated field names: column1, column_1, col1, field1, item1, attr1
    if re.fullmatch(r"(column|col|field|item|attr|header|row|name|value|data|label|text)[ _-]?\d+", s):
        return True

    # repeated symbol strings
    if re.fullmatch(r"[-*#.@]{3,}", s):
        return True

    return False


def unique_ratio(values):
    """
    Unique ratio over non-empty normalized strings.
    """
    vals = [normalize_text(v) for v in values if normalize_text(v) != ""]
    if not vals:
        return None
    return len(set(vals)) / len(vals)


def numeric_ratio(values):
    """
    Numeric-like ratio over non-empty normalized strings.
    """
    vals = [normalize_text(v) for v in values if normalize_text(v) != ""]
    if not vals:
        return None
    return sum(is_numeric_like(v) for v in vals) / len(vals)


def placeholder_ratio(values):
    """
    Placeholder-like ratio over non-empty values.
    """
    vals = [v for v in values if normalize_text(v) != ""]
    if not vals:
        return None
    return sum(is_placeholder_like(v) for v in vals) / len(vals)


def mean_string_length(values):
    vals = [normalize_text(v) for v in values if normalize_text(v) != ""]
    if not vals:
        return None
    return sum(len(v) for v in vals) / len(vals)


def find_effective_region(ws):
    """
    Find the minimal bounding rectangle covering all non-empty cells.
    Returns (min_row, max_row, min_col, max_col), 1-based indexing,
    or None if no non-empty cells exist.
    """
    min_row = None
    max_row = None
    min_col = None
    max_col = None

    for row in ws.iter_rows():
        for cell in row:
            if not is_empty_cell(cell.value):
                r, c = cell.row, cell.column
                if min_row is None or r < min_row:
                    min_row = r
                if max_row is None or r > max_row:
                    max_row = r
                if min_col is None or c < min_col:
                    min_col = c
                if max_col is None or c > max_col:
                    max_col = c

    if min_row is None:
        return None

    return min_row, max_row, min_col, max_col


def extract_region_values(ws, region):
    """
    Extract all values from the effective region into a 2D Python list.
    """
    min_row, max_row, min_col, max_col = region
    data = []

    for row in ws.iter_rows(
        min_row=min_row,
        max_row=max_row,
        min_col=min_col,
        max_col=max_col,
        values_only=True
    ):
        data.append(list(row))

    return data


def evaluate_worksheet(
    ws,
    max_rows=DEFAULT_MAX_ROWS,
    max_cols=DEFAULT_MAX_COLS,
    min_rows=DEFAULT_MIN_ROWS,
    min_cols=DEFAULT_MIN_COLS,
    min_fill_ratio=DEFAULT_MIN_FILL_RATIO,
    max_missing_noise_ratio=DEFAULT_MAX_MISSING_NOISE_RATIO,
    max_placeholder_ratio=DEFAULT_MAX_PLACEHOLDER_RATIO,
    min_header_unique_ratio=DEFAULT_MIN_HEADER_UNIQUE_RATIO,
    max_numeric_header_ratio=DEFAULT_MAX_NUMERIC_HEADER_RATIO,
):
    """
    Compute worksheet metadata and keep/drop decision.
    """
    metadata = {
        "sheet_name": ws.title,
        "has_effective_region": False,
        "effective_min_row": None,
        "effective_max_row": None,
        "effective_min_col": None,
        "effective_max_col": None,
        "n_rows": 0,
        "n_cols": 0,
        "area": 0,
        "non_empty_cells": 0,
        "fill_ratio": None,
        "missing_noise_cells": 0,
        "missing_noise_ratio": None,
        "header_row_nonempty": 0,
        "header_col_nonempty": 0,
        "header_row_placeholder_ratio": None,
        "header_col_placeholder_ratio": None,
        "header_row_unique_ratio": None,
        "header_col_unique_ratio": None,
        "header_row_numeric_ratio": None,
        "header_col_numeric_ratio": None,
        "header_row_mean_len": None,
        "header_col_mean_len": None,
        "keep_sheet": False,
        "drop_reasons": "",
    }

    reasons = []

    region = find_effective_region(ws)
    if region is None:
        reasons.append("no_effective_region")
        metadata["drop_reasons"] = ";".join(reasons)
        return metadata

    metadata["has_effective_region"] = True
    metadata["effective_min_row"], metadata["effective_max_row"], metadata["effective_min_col"], metadata["effective_max_col"] = region

    values_2d = extract_region_values(ws, region)
    R = len(values_2d)
    C = len(values_2d[0]) if R > 0 else 0
    area = R * C

    metadata["n_rows"] = R
    metadata["n_cols"] = C
    metadata["area"] = area

    # Size rules
    if R < min_rows:
        reasons.append("too_few_rows")
    if C < min_cols:
        reasons.append("too_few_cols")
    if R > max_rows:
        reasons.append("too_many_rows")
    if C > max_cols:
        reasons.append("too_many_cols")

    if area < 9:
        reasons.append("area_lt_9")

    # Cell statistics
    flat_values = [v for row in values_2d for v in row]
    non_empty_cells = sum(not is_empty_cell(v) for v in flat_values)
    metadata["non_empty_cells"] = non_empty_cells

    fill_ratio = non_empty_cells / area if area > 0 else 0.0
    metadata["fill_ratio"] = fill_ratio
    if fill_ratio < min_fill_ratio:
        reasons.append("low_fill_ratio")

    missing_noise_cells = sum(
        is_missing_marker(v) or is_garbled(v)
        for v in flat_values
    )
    metadata["missing_noise_cells"] = missing_noise_cells

    missing_noise_ratio = missing_noise_cells / area if area > 0 else 0.0
    metadata["missing_noise_ratio"] = missing_noise_ratio
    if missing_noise_ratio > max_missing_noise_ratio:
        reasons.append("high_missing_noise_ratio")

    # Header candidates
    header_row = values_2d[0] if R >= 1 else []
    header_col = [row[0] for row in values_2d] if C >= 1 else []

    # Exclude top-left cell from both header stats to avoid double counting
    if header_row:
        header_row_eval = header_row[1:] if len(header_row) > 1 else []
    else:
        header_row_eval = []

    if header_col:
        header_col_eval = header_col[1:] if len(header_col) > 1 else []
    else:
        header_col_eval = []

    header_row_nonempty = sum(normalize_text(v) != "" for v in header_row_eval)
    header_col_nonempty = sum(normalize_text(v) != "" for v in header_col_eval)

    metadata["header_row_nonempty"] = header_row_nonempty
    metadata["header_col_nonempty"] = header_col_nonempty

    metadata["header_row_placeholder_ratio"] = placeholder_ratio(header_row_eval)
    metadata["header_col_placeholder_ratio"] = placeholder_ratio(header_col_eval)
    metadata["header_row_unique_ratio"] = unique_ratio(header_row_eval)
    metadata["header_col_unique_ratio"] = unique_ratio(header_col_eval)
    metadata["header_row_numeric_ratio"] = numeric_ratio(header_row_eval)
    metadata["header_col_numeric_ratio"] = numeric_ratio(header_col_eval)
    metadata["header_row_mean_len"] = mean_string_length(header_row_eval)
    metadata["header_col_mean_len"] = mean_string_length(header_col_eval)

    # Apply header rules only if there are at least 3 non-empty candidate headers
    if header_row_nonempty >= 3:
        pr = metadata["header_row_placeholder_ratio"]
        ur = metadata["header_row_unique_ratio"]
        nr = metadata["header_row_numeric_ratio"]

        if pr is not None and pr > max_placeholder_ratio:
            reasons.append("bad_column_headers_placeholder")
        if ur is not None and ur < min_header_unique_ratio:
            reasons.append("bad_column_headers_low_unique")
        if nr is not None and nr > max_numeric_header_ratio:
            reasons.append("bad_column_headers_too_numeric")

    if header_col_nonempty >= 3:
        pr = metadata["header_col_placeholder_ratio"]
        ur = metadata["header_col_unique_ratio"]
        nr = metadata["header_col_numeric_ratio"]

        if pr is not None and pr > max_placeholder_ratio:
            reasons.append("bad_row_headers_placeholder")
        if ur is not None and ur < min_header_unique_ratio:
            reasons.append("bad_row_headers_low_unique")
        if nr is not None and nr > max_numeric_header_ratio:
            reasons.append("bad_row_headers_too_numeric")

    metadata["keep_sheet"] = len(reasons) == 0
    metadata["drop_reasons"] = ";".join(reasons)

    return metadata


def process_workbook(
    xlsx_path,
    **kwargs
):
    """
    Evaluate all worksheets in one workbook.
    Returns:
        sheet_records: list of worksheet metadata dict
        file_summary: dict
    """
    sheet_records = []

    try:
        wb = load_workbook(
            filename=xlsx_path,
            read_only=True,
            data_only=True
        )
    except Exception as e:
        return [], {
            "file_path": str(xlsx_path),
            "file_name": Path(xlsx_path).name,
            "num_sheets_total": 0,
            "num_sheets_kept": 0,
            "keep_file": False,
            "drop_reason": f"workbook_read_error:{type(e).__name__}"
        }

    for ws in wb.worksheets:
        meta = evaluate_worksheet(ws, **kwargs)
        meta["file_path"] = str(xlsx_path)
        meta["file_name"] = Path(xlsx_path).name
        sheet_records.append(meta)

    num_sheets_total = len(sheet_records)
    num_sheets_kept = sum(r["keep_sheet"] for r in sheet_records)

    # File-level rule:
    # Keep workbook if at least one worksheet is valid
    keep_file = num_sheets_kept > 0

    if keep_file:
        drop_reason = ""
    else:
        if num_sheets_total == 0:
            drop_reason = "no_worksheets"
        else:
            drop_reason = "all_worksheets_filtered"

    file_summary = {
        "file_path": str(xlsx_path),
        "file_name": Path(xlsx_path).name,
        "num_sheets_total": num_sheets_total,
        "num_sheets_kept": num_sheets_kept,
        "keep_file": keep_file,
        "drop_reason": drop_reason,
    }

    try:
        wb.close()
    except Exception:
        pass

    return sheet_records, file_summary


def find_xlsx_files(input_dir):
    """
    Recursively find all .xlsx files, excluding temporary Excel lock files.
    """
    input_dir = Path(input_dir)
    files = []
    for p in input_dir.rglob("*.xlsx"):
        if p.name.startswith("~$"):
            continue
        files.append(p)
    return sorted(files)


def write_txt_lines(path, lines):
    with open(path, "w", encoding="utf-8") as f:
        for line in lines:
            f.write(str(line) + "\n")


def main():
    parser = argparse.ArgumentParser(description="Filter XLSX dataset by worksheet quality rules.")
    parser.add_argument("--input_dir", type=str, required=True, help="Input folder containing .xlsx files")
    parser.add_argument("--output_dir", type=str, required=True, help="Output folder for metadata and lists")

    parser.add_argument("--max_rows", type=int, default=DEFAULT_MAX_ROWS)
    parser.add_argument("--max_cols", type=int, default=DEFAULT_MAX_COLS)
    parser.add_argument("--min_rows", type=int, default=DEFAULT_MIN_ROWS)
    parser.add_argument("--min_cols", type=int, default=DEFAULT_MIN_COLS)
    parser.add_argument("--min_fill_ratio", type=float, default=DEFAULT_MIN_FILL_RATIO)
    parser.add_argument("--max_missing_noise_ratio", type=float, default=DEFAULT_MAX_MISSING_NOISE_RATIO)
    parser.add_argument("--max_placeholder_ratio", type=float, default=DEFAULT_MAX_PLACEHOLDER_RATIO)
    parser.add_argument("--min_header_unique_ratio", type=float, default=DEFAULT_MIN_HEADER_UNIQUE_RATIO)
    parser.add_argument("--max_numeric_header_ratio", type=float, default=DEFAULT_MAX_NUMERIC_HEADER_RATIO)

    args = parser.parse_args()

    output_dir = Path(args.output_dir)
    output_dir.mkdir(parents=True, exist_ok=True)

    xlsx_files = find_xlsx_files(args.input_dir)
    print(f"Found {len(xlsx_files)} xlsx files.")

    worksheet_records = []
    file_records = []

    for i, xlsx_path in enumerate(xlsx_files, 1):
        if i % 100 == 0 or i == 1:
            print(f"[{i}/{len(xlsx_files)}] Processing: {xlsx_path}")

        sheet_records, file_summary = process_workbook(
            xlsx_path,
            max_rows=args.max_rows,
            max_cols=args.max_cols,
            min_rows=args.min_rows,
            min_cols=args.min_cols,
            min_fill_ratio=args.min_fill_ratio,
            max_missing_noise_ratio=args.max_missing_noise_ratio,
            max_placeholder_ratio=args.max_placeholder_ratio,
            min_header_unique_ratio=args.min_header_unique_ratio,
            max_numeric_header_ratio=args.max_numeric_header_ratio,
        )

        worksheet_records.extend(sheet_records)
        file_records.append(file_summary)

    # Save worksheet metadata
    worksheet_df = pd.DataFrame(worksheet_records)
    worksheet_csv = output_dir / "worksheet_metadata.csv"
    worksheet_df.to_csv(worksheet_csv, index=False, encoding="utf-8-sig")

    # Save file summary
    file_df = pd.DataFrame(file_records)
    file_csv = output_dir / "file_summary.csv"
    file_df.to_csv(file_csv, index=False, encoding="utf-8-sig")

    kept_files = file_df.loc[file_df["keep_file"] == True, "file_path"].tolist()
    dropped_files = file_df.loc[file_df["keep_file"] == False, "file_path"].tolist()

    write_txt_lines(output_dir / "kept_files.txt", kept_files)
    write_txt_lines(output_dir / "dropped_files.txt", dropped_files)

    # Save simple statistics
    stats = {
        "num_xlsx_total": len(file_df),
        "num_xlsx_kept": int(file_df["keep_file"].sum()) if len(file_df) > 0 else 0,
        "num_xlsx_dropped": int((~file_df["keep_file"]).sum()) if len(file_df) > 0 else 0,
        "num_worksheets_total": len(worksheet_df),
        "num_worksheets_kept": int(worksheet_df["keep_sheet"].sum()) if len(worksheet_df) > 0 else 0,
        "num_worksheets_dropped": int((~worksheet_df["keep_sheet"]).sum()) if len(worksheet_df) > 0 else 0,
    }

    stats_path = output_dir / "stats_summary.csv"
    pd.DataFrame([stats]).to_csv(stats_path, index=False, encoding="utf-8-sig")

    print("\nDone.")
    print(f"Worksheet metadata saved to: {worksheet_csv}")
    print(f"File summary saved to:      {file_csv}")
    print(f"Stats summary saved to:     {stats_path}")
    print(f"Kept files list saved to:   {output_dir / 'kept_files.txt'}")
    print(f"Dropped files list saved to:{output_dir / 'dropped_files.txt'}")


if __name__ == "__main__":
    main()