README.md

metadata

license: mit
task_categories:
  - question-answering
  - text-generation
language:
  - en
tags:
  - theory-of-computation
  - algorithms
  - computer-science
  - homework
  - exercises
size_categories:
  - n<1K
pretty_name: LyTOC Benchmark

LyTOC Benchmark Dataset

A curated collection of Theory of Computation and Algorithms homework exercises, extracted from academic PDFs using OCR and structured for machine learning evaluation.

🔗 Links:

• 📦 HuggingFace Dataset
• 💻 GitHub Repository

Dataset Description

Dataset Summary

The LyTOC (Logic and Theory of Computation) Benchmark contains 27 carefully extracted exercises from 9 homework assignments covering fundamental topics in theoretical computer science. Each exercise is preserved with its original LaTeX mathematical notation, making it suitable for evaluating language models on formal reasoning tasks.

Key Features:

• 27 exercises across 9 homework assignments
• Topics: automata theory, complexity theory, Turing machines, formal languages, algorithm analysis
• LaTeX mathematical notation preserved
• Structured with exercise numbers
• Clean extraction with OCR post-processing

Supported Tasks

• Question Answering: Answer theoretical computer science questions
• Mathematical Reasoning: Solve problems involving formal proofs and mathematical notation
• Text Generation: Generate solutions to computational theory problems
• Educational Assessment: Evaluate understanding of CS theory concepts

Languages

• English (en)

Dataset Structure

Data Instances

Each instance represents a single exercise:

{
  "homework": "hw1",
  "exercise_number": "3",
  "content": "Let $\\Sigma = \\{0, 1\\}$. Let language\n\n$$L = \\{w \\in \\{0, 1\\}^* : w \\text{ has an unequal number of 0's and 1's}\\}.$$\n\nProve $L^* = \\Sigma^*$.",
  "full_id": "hw1_ex3"
}

Data Fields

• homework (string): Homework identifier (e.g., "hw1", "hw2", "hw13")
• exercise_number (string): Exercise number within the homework (e.g., "1", "2", "3")
• content (string): Full exercise text including LaTeX mathematical notation
• full_id (string): Unique identifier for the exercise (e.g., "hw1_ex3", "hw2_ex3_1")

Data Splits

The dataset consists of a single split containing all 27 exercises.

Dataset Statistics

• Total Exercises: 27
• Homeworks: 9 (hw1, hw2, hw3, hw5, hw6, hw9, hw10, hw11, hw13)
• Average Content Length: ~200-500 characters per exercise

Topic Distribution

The exercises cover the following topics:

• Asymptotic Analysis: Big-O notation, growth rates
• Finite Automata: DFA, NFA, regular expressions
• Formal Languages: Regular languages, context-free languages
• Turing Machines: Decidability, computability
• Complexity Theory: P, NP, NP-completeness, reductions
• Algorithm Design: Time complexity, space complexity

Dataset Creation

Source Data

The dataset was created from homework assignments in a Theory of Computation and Algorithms course.

Data Collection

• Source: Academic homework PDFs (9 files)
• Extraction Method: SimpleTex OCR API
• Processing: Automated regex-based exercise splitting
• Quality Control: Manual verification of extraction accuracy

Data Processing Pipeline

1. PDF to Image: Convert each PDF page to high-resolution images
2. OCR Processing: Extract text using SimpleTex OCR API
3. Punctuation Normalization: Convert Chinese punctuation to English equivalents
4. Exercise Splitting: Use regex patterns to identify exercise boundaries
5. Metadata Generation: Create unique identifiers and structure data

Annotations

The dataset does not include solutions or annotations. It contains only problem statements as extracted from the source materials.

Considerations for Using the Data

Recommended Uses

• Evaluating language models on formal reasoning tasks
• Training models for mathematical problem understanding
• Benchmarking CS theory knowledge in AI systems
• Educational tool development for computer science

Limitations

• No Solutions: The dataset contains only problem statements, not solutions
• OCR Artifacts: Some mathematical notation may have minor OCR errors
• Limited Scope: Covers specific topics in theory of computation and algorithms
• No Visual Content: Diagrams and figures from PDFs are not included
• Language: English only

Ethical Considerations

This dataset is intended for educational and research purposes. Users should:

• Respect academic integrity when using for educational purposes
• Not use for automated homework completion systems
• Cite appropriately when using in research

Additional Information

Licensing Information

This dataset is released under the MIT License.

Citation Information

If you use this dataset in your research, please cite:

@misc{lytoc-benchmark-2025,
  title={LyTOC Benchmark: Theory of Computation and Algorithms Exercise Dataset},
  author={LyTOC Contributors},
  year={2025},
  howpublished={\\url{https://huggingface.co/datasets/lytoc-benchmark}}
}

Dataset Curators

Dataset created using:

• SimpleTex OCR API for PDF extraction
• Custom Python scripts for data processing
• Claude Code for automation and quality assurance

Contact

For questions or issues regarding this dataset, please open an issue on the dataset repository.

Usage Example

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("Zecyel/LyTOC")

# Access an exercise
exercise = dataset['train'][0]
print(f"Exercise ID: {exercise['full_id']}")
print(f"Content: {exercise['content']}")

# Filter by homework
hw1_exercises = [ex for ex in dataset['train'] if ex['homework'] == 'hw1']
print(f"Homework 1 has {len(hw1_exercises)} exercises")

Version History

• v1.0.0 (2025-12-30): Initial release with 27 exercises from 9 homework assignments