USAMO/segment_script/segment_2025.py

# -----------------------------------------------------------------------------
# Author: Jiawei Liu
# Date: 2025-10-29
# -----------------------------------------------------------------------------
import json
import re
from pathlib import Path

from rapidfuzz import fuzz


def clean_text(text: str):
    text = re.sub(
        r"\nAvailable online at.+?\.\s\s$", "", text, flags=re.DOTALL | re.MULTILINE
    )
    return text


def extract_problems(markdown_text: str):
    problems = {}

    # 1. Find the block of text containing the problems
    # Block starts with '## Problems'
    # and ends before '## \(S 1\) Solutions to Day 1'
    problems_section = re.search(
        r"^## Problems\s*$(.*?)^##.*Solutions to Day 1",
        markdown_text,
        re.DOTALL | re.MULTILINE,
    )

    if not problems_section:
        print("  - Not found '## Problems' section.")
        return problems

    problems_block = problems_section.group(1)

    # 2. Split the block into individual problems
    # Each problem starts with a number followed by a dot and space (e.g., '1. ')
    matchs = list(re.finditer(r"^(\d+)\.\s+", problems_block, flags=re.MULTILINE))

    # Split the text into parts using the problem numbers as delimiters
    for i, m in enumerate(matchs):
        problem_label = m.group(1)
        problem = problems_block[
            m.end() : matchs[i + 1].start()
            if i + 1 < len(matchs)
            else len(problems_block)
        ]

        problems[problem_label] = (problem.strip(), m.group())

    print(f"  - Extracted {len(problems)} problems.")

    return problems


def extract_solutions(markdown_text):
    solutions = {}

    # 1. Find the block of text containing the solutions
    # Block starts with '## \(S 1\) Solutions to Day 1' and end of document
    solutions_section = re.search(
        r"^##.*?Solutions to Day 1\s*$(.*)",
        markdown_text,
        re.DOTALL | re.MULTILINE,
    )

    if not solutions_section:
        print("  - Not found Solutions section.")
        return solutions

    solutions_block = solutions_section.group(1)

    ## \(\S 1.1\) USAMO 2025/1, proposed by John Berman  
    # 2. Split the block into individual solutions
    matchs = list(
        re.finditer(r"^##.*?USAMO\s*?\d{4}\/(\d+).*?\n$", solutions_block, flags=re.MULTILINE)
    )

    # Split the text into parts using the solution numbers as delimiters
    for i, m in enumerate(matchs):
        problem_label = m.group(1)
        solution = solutions_block[
            m.end() : matchs[i + 1].start()
            if i + 1 < len(matchs)
            else len(solutions_block)
        ]

        solutions[problem_label] = (solution.strip(), m.group())

    print(f"  - Extracted {len(solutions)} solutions.")

    return solutions


def join(problems: dict, solutions: dict):
    pairs = []

    for problem_label, (problem, p_match) in problems.items():
        solution, s_match = solutions.get(problem_label)

        # Clean solution by removing the part that overlaps with the problem statement
        problem_align = fuzz.partial_ratio_alignment(solution, problem)
        solution = solution.replace(
            solution[problem_align.src_start : problem_align.src_end], ""
        )
        solution = re.sub(
            r"^\s*## Problem statement", "", solution, flags=re.IGNORECASE
        ).strip()

        if not solution:
            print(f"  - Warning: No solution found for problem {problem_label}.")
        pairs.append((problem, solution, problem_label, p_match, s_match))

    return pairs


def write_pairs(output_file: Path, pairs: list, year: str, project_root: Path):
    output_jsonl_text = ""
    for problem, solution, problem_label, p_match, s_match in pairs:
        output_jsonl_text += (
            json.dumps(
                {
                    "year": year,
                    "tier": "T1",
                    "problem_label": problem_label,
                    "problem_type": None,
                    "exam": "USAMO",
                    "problem": problem,
                    "solution": solution,
                    "metadata": {
                        "resource_path": output_file.relative_to(
                            project_root
                        ).as_posix(),
                        "problem_match": p_match,
                        "solution_match": s_match,
                    },
                },
                ensure_ascii=False,
            )
            + "\n"
        )

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


if __name__ == "__main__":
    compet_base_path = Path(__file__).resolve().parent.parent
    compet_md_path = compet_base_path / "md"
    seg_output_path = compet_base_path / "segmented"
    project_root = compet_base_path.parent

    num_problems = 0
    num_solutions = 0

    md_files = [
        _
        for _ in compet_md_path.glob("**/*.md")
        if int(re.search(r"\d{4}", _.as_posix()).group()) in [2025]
    ]
    for md_file in md_files:
        print(f"Processing {md_file}...")
        output_file = seg_output_path / md_file.relative_to(compet_md_path).with_suffix(
            ".jsonl"
        )
        output_file.parent.mkdir(parents=True, exist_ok=True)

        # Read the markdown file
        markdown_text = "\n" + md_file.read_text(encoding="utf-8")
        markdown_text = clean_text(markdown_text)

        problems = extract_problems(markdown_text)
        solutions = extract_solutions(markdown_text)

        num_problems += len(problems)
        num_solutions += len(solutions)

        pairs = join(problems, solutions)

        year = re.search(r"\d{4}", output_file.stem).group()

        write_pairs(output_file, pairs, year, project_root)

        print()

    print(f"Total problems extracted: {num_problems}")
    print(f"Total solutions extracted: {num_solutions}")