ИСТИНА |
Войти в систему Регистрация |
|
ИСТИНА ИНХС РАН |
||
The Interactive Visual Explorer (InVEx) application is designed as a visual analytics tool for Big Data analysis. Visual analytics is an integral approach to data analysis, combining methods of the intellectual data analysis with the advanced interactive visualization. Dealing with large amounts of multidimensional data implies grouping or clustering this data in a variety of ways. One of the goals of the InVEx is to process large data samples by decreasing their level of detail (LoD), which is achieved with various grouping approaches. Its current implementation uses MiniBatchKMeans algorithm (Scikit-learn library), providing high performance clustering of the vast samples of data objects. The core idea of the LoD method is to split the initial data sample into groups of objects, sharing similar parameters values, and represent these groups as an aggregated objects for theirs further visualization as spheres on 3D scene. Thereby, the application provides ability of nested (hierarchical) groupings, so aggregated data objects created from the initial data sample (or the previous step of grouping) could be grouped again, while giving the possibility to explore objects within the group. K-means clusterization can be used only for numerical continuous parameters, while each data sample may contain parameters of different data types and statistical measurements like ratio, interval, ordinal and nominal. And often there is a need to group data by various combinations of these parameters. Proposed approach of nested intellectual data grouping and clusterization for the InVEx includes clustering methods for mixed data as well as flexible grouping by different parameters, providing the exploration of data from the lowest to the highest level of details. The results of grouping and clusterization are visualized using interactive 3D scene and parallel coordinates, allowing users to get insights of data, to explore hidden correlations and trends of parameters.