The Khiops suite builds supervised models (classifiers and regressors) as well as basic and advanced unsupervised models (hierarchical coclusterings) for exploratory analysis. The suite also provides visualization tools to analyze and interpret the resulting models.
Khiops automatizes many time-consuming tasks including:
- Feature engineering for large relational datasets
- Discretization of numerical features and grouping of categorical ones
- Decision tree creation
- Feature selection in supervised models
Khiops achieves an excellent compromise between accuracy, speed of learning and deployment, robustness and explainability. Its modeling algorithms are based on information theoretic principles and have been object of many peer-reviewed publications.
Khiops is implemented in C++ and makes the best use of your CPU cores, RAM and disk capacity. Khiops algorithms scale well with very large datasets by efficiently using disk storage when they do not fit entirely in RAM and by using multiple CPU cores.
Khiops is a production-level ML suite that has been extensively used in real world applications: At Orange it is currently used for churn detection, fraud detection, sentiment analysis, appetency modeling and many others.
Khiops runs natively on various Linux distributions, Windows systems as well as clusters. Several integration enablers are available, such as pyKhiops a complete Python API.