Upload README.md with huggingface_hub

README.md

@@ -1,146 +1,193 @@
 ---
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-dataset_info:
-  features:
-  - name: homework
-    dtype: string
-  - name: exercise_number
-    dtype: string
-  - name: sub_problem
-    dtype: string
-  - name: content
-    dtype: string
-  - name: full_id
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 6972
-    num_examples: 28
-  download_size: 6910
-  dataset_size: 6972
+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
 
-This repository contains a benchmark dataset extracted from homework PDFs using SimpleTex OCR API.
+# LyTOC Benchmark Dataset
 
-## Features
+A curated collection of Theory of Computation and Algorithms homework exercises, extracted from academic PDFs using OCR and structured for machine learning evaluation.
 
-- Automated PDF content extraction using SimpleTex OCR API
-- Structured benchmark dataset creation
-- Multiple export formats (JSON, JSONL, HuggingFace Dataset)
-- Easy upload to HuggingFace Hub
-- Interactive pipeline runner
+## Dataset Description
 
-## Quick Start
+### Dataset Summary
 
-### 1. Install Dependencies
+The LyTOC (Logic and Theory of Computation) Benchmark contains 28 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.
 
-```bash
-pip install -r requirements.txt
-```
+**Key Features:**
+- 28 exercises across 9 homework assignments
+- Topics: automata theory, complexity theory, Turing machines, formal languages, algorithm analysis
+- LaTeX mathematical notation preserved
+- Structured with exercise numbers and sub-problems
+- Clean extraction with OCR post-processing
 
-### 2. Set Up API Keys
+### Supported Tasks
 
-```bash
-cp .env.example .env
-# Edit .env and add your API keys:
-# - OCR_UAT: Get from https://simpletex.cn
-# - HF_TOKEN: Get from https://huggingface.co/settings/tokens
-```
+- **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
 
-### 3. Run the Pipeline
+### Languages
 
-**Option A: Interactive Pipeline** (Recommended)
-```bash
-python run_pipeline.py
-```
+- English (en)
 
-**Option B: Manual Steps**
-```bash
-# Extract PDFs
-python extract_pdfs.py
+## Dataset Structure
+
+### Data Instances
 
-# Create benchmark dataset
-python create_benchmark.py
+Each instance represents a single exercise or sub-problem:
 
-# Upload to HuggingFace
-python upload_to_hf.py <username/repo-name> [--private]
+```json
+{
+  "homework": "hw1",
+  "exercise_number": "3",
+  "sub_problem": null,
+  "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"
+}
 ```
 
-## Project Structure
+### Data Fields
 
-```
-lytoc/
-├── raw/                          # Source PDF files
-├── parsed_data/                  # Extracted markdown content
-│   ├── hw*.md                   # Individual parsed files
-│   └── extraction_metadata.json # Extraction status
-├── benchmark_dataset/            # Final benchmark dataset
-│   ├── dataset.json             # JSON format
-│   ├── dataset.jsonl            # JSONL format
-│   └── huggingface_dataset/     # HuggingFace format
-├── extract_pdfs.py              # PDF extraction script
-├── create_benchmark.py          # Benchmark creation script
-├── upload_to_hf.py              # HuggingFace upload script
-├── run_pipeline.py              # Interactive pipeline runner
-└── DATASET_CARD.md              # Dataset documentation
-```
+- `homework` (string): Homework identifier (e.g., "hw1", "hw2", "hw13")
+- `exercise_number` (string): Exercise number within the homework (e.g., "1", "2", "3")
+- `sub_problem` (string or null): Sub-problem identifier if the exercise has multiple parts (e.g., "1", "2")
+- `content` (string): Full exercise text including LaTeX mathematical notation
+- `full_id` (string): Unique identifier for the exercise (e.g., "hw1_ex3", "hw2_ex3_1")
 
-## Dataset Structure
+### Data Splits
 
-Each problem in the dataset contains:
-- `homework`: Homework identifier (e.g., "hw1", "hw2")
-- `problem_number`: Problem number within the homework
-- `content`: Full problem text and description
-- `full_id`: Unique identifier (e.g., "hw1_problem1")
+The dataset consists of a single split containing all 28 exercises.
 
-Example:
-```json
-{
-  "homework": "hw1",
-  "problem_number": "1",
-  "content": "Problem statement...",
-  "full_id": "hw1_problem1"
+## Dataset Statistics
+
+- **Total Exercises**: 28
+- **Homeworks**: 9 (hw1, hw2, hw3, hw5, hw6, hw9, hw10, hw11, hw13)
+- **Exercises with Sub-problems**: 2
+- **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. **Sub-problem Detection**: Identify and separate sub-problems within exercises
+6. **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:
+
+```bibtex
+@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}}
 }
 ```
 
-## Scripts
+### Dataset Curators
 
-### extract_pdfs.py
-Extracts content from PDF files in the `raw/` directory using SimpleTex OCR API.
-- Converts PDFs to markdown format via OCR
-- Saves parsed content to `parsed_data/`
-- Creates extraction metadata
+Dataset created using:
+- SimpleTex OCR API for PDF extraction
+- Custom Python scripts for data processing
+- Claude Code for automation and quality assurance
 
-### create_benchmark.py
-Processes parsed content into a structured benchmark dataset.
-- Parses markdown content to extract individual problems
-- Creates dataset in multiple formats
-- Generates statistics and sample output
+### Contact
 
-### upload_to_hf.py
-Uploads the benchmark dataset to HuggingFace Hub.
-```bash
-python upload_to_hf.py username/repo-name [--private]
-```
+For questions or issues regarding this dataset, please open an issue on the dataset repository.
 
-### run_pipeline.py
-Interactive script that runs the complete pipeline with user prompts.
+## Usage Example
 
-## Requirements
+```python
+from datasets import load_dataset
 
-- Python 3.8+
-- SimpleTex API token (OCR_UAT)
-- HuggingFace account and token (for upload)
+# Load the dataset
+dataset = load_dataset("lytoc-benchmark")
 
-## License
+# Access an exercise
+exercise = dataset['train'][0]
+print(f"Exercise ID: {exercise['full_id']}")
+print(f"Content: {exercise['content']}")
 
-MIT License - See LICENSE file for details
+# Filter by homework
+hw1_exercises = [ex for ex in dataset['train'] if ex['homework'] == 'hw1']
+print(f"Homework 1 has {len(hw1_exercises)} exercises")
+```
 
-## Contributing
+## Version History
 
-Contributions are welcome! Please feel free to submit a Pull Request.
+- **v1.0.0** (2025-12-30): Initial release with 28 exercises from 9 homework assignments