This project provides a fully automated deep learning framework for multi-label classification across various data modalities, including images, video, audio, text, and tabular data. It aims to eliminate manual intervention in data preprocessing, feature engineering, model selection, and hyperparameter tuning, enabling rapid development of high-performance classifiers. The target audience includes researchers and practitioners seeking to quickly deploy robust classification models for diverse real-world problems.

How It Works

AutoDL employs a unified algorithmic pipeline that intelligently handles data, features, and models. It supports a wide range of traditional and deep learning models, from SVM and XGBoost to ResNet, BERT, and GRU. The framework is designed for speed, capable of producing competitive results in as little as 10 seconds, with real-time feedback on model performance. This approach addresses common challenges like resource constraints, data imbalance, small datasets, and complex feature engineering.

Quick Start & Requirements

• Installation: pip install autodl-gpu (or autodl-cpu).
• Prerequisites: Python >= 3.5, PyTorch 1.3.1, TensorFlow 1.15. For GPU support, CUDA 10 and cuDNN 7.5 are recommended. Docker images are available for CPU and GPU.
• Setup: Local testing involves cloning the repository and running python run_local_test.py. Detailed examples for various data types are provided.
• Resources: The project includes scripts to download public datasets, with sizes ranging from KB to GB.

Highlighted Details

• Winner of the AutoDL Challenge @ NeurIPS, achieving top rankings across multiple datasets.
• Supports arbitrary multi-label classification tasks (binary, multi-class, multi-label).
• Demonstrates strong performance across 24 offline and 15 online datasets in diverse domains.
• Offers real-time performance feedback via HTML reports.

Maintenance & Community

The project encourages community contributions via issues and pull requests. Links to community channels (WeChat) are provided.

Licensing & Compatibility

Licensed under the Apache License 2.0. This permissive license generally allows for commercial use and integration into closed-source projects.

Limitations & Caveats

The project specifies compatibility with Python 3.6+, PyTorch 1.3.1, and TensorFlow 1.15, which are older versions. Users may encounter compatibility issues with newer environments. The installation instructions for Windows include specific older versions of CUDA, cuDNN, and Miniconda.