HarvardMIT/segment_script/segment.py

import re
import json

from tqdm import tqdm
from loguru import logger

from pathlib import Path
from typing import Tuple, List
from dataclasses import dataclass


project_root = Path(__file__).parent.parent.parent


@dataclass
class Problem:
    match: re.Match


@dataclass
class Solution:
    match: re.Match


def clean_text(text: str) -> str:
    text = text.replace(
        'For a discussion, see\nW. Morris and V. Soltan. The Erdős-Szekeres Problem on Points in Convex Postion-A Survey, Bulletin of the American Math Monthly. 37 (2000), 437-458.\n\nThis article is available at\nhttp://www.ams.org/bull/2000-37-04/S0273-0979-00-00877-6/home.html.\nIf $N(7)=33$, the highest sure score on this problem would be $32-6=26$. It is not known whether there exist arbitrarily large sets of points that will fool the graders.\n\n## The unexamined life is not worth living.',
        ''
    )
    text = text.replace(
        '- Bishop: This piece can move any number of squares diagonally if there are no other pieces along its path.\n- Rook: This piece can move any number of squares either vertically or horizontally if there are no other pieces along its path\n- Knight: This piece can move either two squares along a row and one square along a column or two squares along a column and one square along a row.\n- King: This piece can move to any open adjacent square (including diagonally).',
        ''
    )
    return text


def find_problem_with_solution(
    text: str,
    problem_parttern: re.Pattern,
    solution_pattern: re.Pattern
) -> int:
    """
    Find the problem with solution start position in the text.
    Args:
        text (str): The text to search.
    Returns:
        int: The start position of the problem with solution.
    """
    matchs = list(problem_parttern.finditer(text))

    for index, match in enumerate(matchs):
        section_end_position = matchs[index + 1].start() if index + 1 < len(matchs) else len(text)
        if solution_pattern.search(text[match.start():section_end_position]):
            return match.start()

    return 0


def analyze(text: str) -> Tuple[List[Problem | Solution], int]:
    """
    Analyze the text and return the tags and problem number.
    Args:
        text (str): The markdown text to analyze.
    Returns:
        Tuple[List[Problem | Solution], int]: A tuple containing the tags and problem number.
    """
    problem_pattern = re.compile(r'(?:\n|\n\#+\s+)(?:(\d{1,2})\.\s+(?:problem\:\s*|\$?\[.+?\]\$?)?|problem\s+?(\w+)\s+\[\d+(?:\spoints)?\]|\$([H|M|T]_\{\d+\})\$\.)', re.IGNORECASE)
    solution_pattern = re.compile(r'(?:\n|\n\#+\s+)(?:answer\:|solution(?:\s+\d+)?(?:\:|\.)|Proposed by:.*?\n)\s*', re.IGNORECASE)

    start_position = find_problem_with_solution(text, problem_pattern, solution_pattern)

    tags: List[Problem | Solution] = []
    tags.extend([Problem(x) for x in problem_pattern.finditer(text, start_position)])
    problem_num = len(tags)

    tags.extend([Solution(x) for x in solution_pattern.finditer(text, start_position)])
    tags.sort(key=lambda x: x.match.start())
    return tags, problem_num


def segment(text: str, tags: List[Problem | Solution]) -> List[str]:
    starts = []
    ends = []

    for i in range(len(tags)):
        starts.append(tags[i].match.end())
        if i + 1 < len(tags):
            ends.append(tags[i + 1].match.start())
        else:
            ends.append(len(text))

    return [text[start:end].strip() for start, end in zip(starts, ends)]


def join(tags: List[Problem | Solution], segments: List[str]) -> List[Tuple[str, str, str, str, str]]:
    problem, solution = '', ''
    problem_label, problem_match, solution_match = '', '', ''
    pairs = []

    for tag, segment in zip(tags, segments):
        if isinstance(tag, Problem):
            problem = segment
            problem_match = tag.match.group(0)
            problem_label = tag.match.group(1) or tag.match.group(2) or tag.match.group(3)
        elif problem.strip() != "":
            solution = segment
            solution_match = tag.match.group(0)

            if solution.strip() == "":
                continue

            pairs.append((problem, solution, problem_label, problem_match, solution_match))

    return pairs


def write_pairs(output_file: Path, pairs):
    year = re.search(r'(\d{4})', output_file.stem).group(1)
    problem_type_mapping = {
        "-alg-": "Algebra",
        "-comb-": "Combinatorics",
        "-geo-": "Geometry",
    }

    problem_type = None
    for _k, _v in problem_type_mapping.items():
        if _k in output_file.stem:
            problem_type = _v
            break

    output_jsonl_text = ""
    for problem, solution, problem_label, problem_match, solution_match in pairs:
        output_jsonl_text += json.dumps(
            {
                'year': year,
                'tier': "T4",
                'problem_label': problem_label,
                'problem_type': problem_type,
                "exam": "HMMT", 
                'problem': problem,
                'solution': solution,
                'metadata': {
                    'resource_path': output_file.relative_to(project_root).as_posix(),
                    'problem_match': problem_match,
                    'solution_match': solution_match
                }
            },
            ensure_ascii=False
        ) + '\n'

    output_file.write_text(output_jsonl_text, encoding="utf-8")


def main():
    compet_base_path = Path(__file__).resolve().parent.parent
    compet_md_path = compet_base_path / "md"
    seg_output_path = compet_base_path / "segmented"

    total_problem_count = 0
    total_solution_count = 0

    for hmmt_md in tqdm(list(compet_md_path.glob('**/*.md')), desc='Segmenting'):
        output_file = seg_output_path / hmmt_md.relative_to(compet_md_path).with_suffix('.jsonl')
        output_file.parent.mkdir(parents=True, exist_ok=True)

        text = '\n' + clean_text(hmmt_md.read_text(encoding="utf-8"))

        tags, problem_num = analyze(text)

        segments = segment(text, tags)
        pairs = join(tags, segments)     
        if pairs and problem_num > 0:
            write_pairs(output_file, pairs)

            total_problem_count += problem_num
            total_solution_count += len(pairs)
        else:
            logger.warning(f"No problem found in {hmmt_md}")
        
    logger.info(f"Total problem count: {total_problem_count}")
    logger.info(f"Total solution count: {total_solution_count}")


if __name__ == '__main__':
    main()