Core Features

• AutoML：Automatic machine learning algorithm selection and parameter tuning.

• AutoFE: Automatic advanced feature generation, evaluation and selection.

• AutoML and AutoFE for time series machine learning tasks.

• Machine learning analysis report generation, reproduce without ChangTianML, autoML experiment records.

• Model Serving: Make predictions, continual learning with new data.

Our platform can build ML models without any coding. Simply upload the dataset(currently tabular classification or regression task) , set feature and label column , set autoML time budget, ChangTianML will automatically build ML pipeline with AutoFE (Automated feature Engineering) and AutoML (automatic model selection and hyperparameter tuning) algorithms, generate new features with strong explanation from original data, and find machine learning models with excellent performance. There is no need to write a single line of code throughout the whole process, which greatly improves the modeling efficiency and lower the knowledge demand to build ML models.

Our AutoFE algorithm defines a large set of operators that imply scientific concepts and meanings in various subjects and industries to generate new features, and then evaluate feature generation steps between optimizing known feature generating space and finding new promising feature generating space. Finally returns an advanced feature group that each has most exclusive expression of data with others and jointly contribute the most gain to ML model. Advanced features will be shown in interpretable formulas, enlightening users with new perspective and new methods for understanding data.

We also supports automatic time series ML. Valuable relations or causations lies deep in patterns and trends within temporal data, so we should not only use current information but also previous data to make final predictions, but how to look back? How long do you look back? How to dig out the advanced features hidden in the timing series? These uncertainties significantly increase the search space for modeling and exploration, and it is more necessary to explore efficiently and intelligently. We extends our automatic machine learning algorithms to time series machine learning to find best time series machine learning models, moreover, we can generate power temporal advanced features, help ML models to build better temporal patterns and trends.

Useful information and findings during autoML will be recorded in Analysis Report, including but not limited to correlation coefficient thermal map analysis, single feature distribution, visual analysis of model effects, and so on. The APIs for generating information above are also provided in downloaded ML model. ML engineers, researchers, data scientists , etc. can easily refine their documents, essays, applications or other work with the help of Analysis Report.