🚀 biggitybiggityO — End-to-End Big-O Time Complexity Classifier

🎥 Quick Loom video

https://www.loom.com/share/f1c8a11cd25a431da79b71eed1b65b2c

📘 Overview

This repository delivers a complete pipeline for building a code time-complexity classifier:

• Collection of real + synthetic complexity-labeled code
• Additional scraping from LeetCode & NeetCode
• Preprocessing, cleaning, merging, and organization of datasets
• Evaluation of multiple pretrained coding models
• Hyperparameter search and QLoRA finetuning
• Model testing & experiment tracking
• REST API + web UI for real-time predictions
• Reproducible environment via Docker

The final model is built on deepseek-coder-1.3b-base for its strong performance-to-size ratio.

🎯 Supported Complexity Classes

• O(1)
• O(log n)
• O(n)
• O(n log n)
• O(n^2)
• O(n^3)
• np (non-polynomial / not predictable)

Additional classes were excluded due to insufficient high-quality training samples.

🏆 Project Outcomes

By the end of this project, the following components were implemented:

• Literature research on complexity-prediction models
• Multi-source dataset creation (real, scraped, synthetic)
• Preprocessing pipelines & dataset merging
• Model selection with structured evaluations
• Hyperparameter search targeting F1-macro
• QLoRA finetuning and testing
• MLflow experiment tracking
• REST API serving
• Frontend UI
• Full CI pipeline
• Dockerized deployment

🔧 Features

• 📥 Automated data scraping
• 🧹 Cleaning & preprocessing pipelines
• 📊 Experiment tracking with MLflow
• 🧠 QLoRA-powered finetuning
• 📝 Time complexity classification API
• 🌐 Frontend for real-time predictions
• 🐳 Dockerfile for fast deployment

📡 Data Sources

• CodeComplex
• Neetcode leetcode solutions
• Leetcode solutions github repo

Installation

Clone the repo and set up the environment:

# 1. Ensure NVIDIA GPU and drivers
nvidia-smi

# 2. Clone the repository
git clone https://github.com/komaksym/biggitybiggityO.git

# 3. Enter into the repository
cd biggitybiggityO

# 4. Build a docker image
docker build -t biggitybiggityo .

# 5. Run the docker image in a new container
docker run --gpus all -p 8000:8000

# 6. Access the web app
Go to http://localhost:8000 in your browser

📂 Directory Structure

biggitybiggityO/
├── app                                      # App itself (API serving and frontend)
│   └── templates                            # Frontend templates
├── data                                     # Everything related to datasets
│   ├── data                                 # Data itself
│   │   ├── codecomplex                      # Data from CodeComplex
│   │   ├── leetcode-parsed                  # Scraped leetcode solutions from github repo
│   │   ├── merges                           # Merges of all of the data sources (except synthetic data)
│   │   ├── neetcode-scraped                 # Scraped leetcode solutions from leetcode
│   │   └── synthetic_data                   # Synthetic data
│   ├── data_experiment                      # Experiment to evaluate performance with synthetic data
│   │   ├── oversampling                     # Oversampling underrepresented classes
│   │   ├── train-mixed_eval-mixed           # Where eval set is a mix of real data and synthetic data
│   │   └── train-synthetic_eval-real        # Where eval set is only real data
│   └── preprocessing_scripts                # Data preprocessing code
│       ├── notebooks                        # Data preprocessing notebooks
│       └── scripts                          # Data preprocessing scripts
├── experiments                              # MLFlow-tracked experiments
├── hyperparameter-search                    # Hyperparameter search results
├── images                                   # Images for README
├── src                                      # Source code
│   ├── eval_competitors                     # Code for evaluating performance of frontier models on test set
│   ├── scraping                             # Code for scraping additional data
│   │   ├── leetcode_solutions               # Scraping from leetcode solutions (github repo)
│   │   └── neetcode                         # Scraping from leetcode solutions (neetcode)
│   └── training                             # Training code
│       ├── code                             # Training source code
│       └── tuned_model_results              # Trained model results for initial model selection
└── tests                                    # Tests
    ├── scraping                             # Scraping tests
    │   └── leetcode                         # Testing leetcode scraping code
    └── training                             # Testing training code
        ├── code                             # Testing training code itself
        └── data                             # Testing data

🤝 Contributing

Contributions are welcome!

• Open an Issue to report bugs or request features
• Submit a Pull Request (PR) for improvements

⭐ Why This Project Matters

This project provides one of the most complete python code complexity datasets available — combining multiple sources. It opens the door for:

• Research on time-complexity prediction
• ML modeling for Big-O classification
• Exploratory analysis of algorithmic patterns
• Reproducible experimentation

👉 If you find this project useful, don’t forget to ⭐ star this repository to support its growth!