The primary aim of AutoML or Automated machine learning is to reduce the skilled human effort required for building machine learning and deep learning models. An AutoML system allows the user to provide the labeled training data as input and receive an optimized model as an output.
Researchers at Tsinghua University have developed an AutoML framework and toolkit specially designed for graph datasets. The driving factor behind this development is that the existing AutoML tools cannot be applied to graphs. Graphs are required everywhere. They are a potent mechanism to represent complex data in a more straightforward, sophisticated manner.
AutoGL(Auto Graph Learning) can handle all stages of graph learning processes, including dataset downloading and management, data preprocessing, model selection and training, hyperparameter tuning and ensembling, etc., automatically. A graph is a structured data-type with nodes and edges. AutoGL supports fully automated machine learning of data on the most common tasks in graph-based machine learning, i.e., graph classification and node classification.
The framework requires PyTorch 1.5.1, Python version 3.6.0, or newer. The AutoGL solvers use auto feature engineering, auto models, hyperparameter optimization, and auto ensemble to solve different graph learning tasks. According to the researchers, their framework will reduce human labor, biases in machine learning graph tasks and serve as a medium for users to implement and test their graph learning methods. The code for this framework is available on Github.