A complete gradient clustering algorithm formed with kernel estimators
The aim of this paper is to provide a gradient clustering algorithm in its complete form, suitable for direct use without requiring a deeper statistical knowledge. The values of all parameters are effectively calculated using optimizing procedures. Moreover, an illustrative analysis of the meaning of particular parameters is shown, followed by the effects resulting from possible modifications with respect to their primarily assigned optimal values. The proposed algorithm does not demand strict assumptions regarding the desired number of clusters, which allows the obtained number to be better suited to a real data structure. Moreover, a feature specific to it is the possibility to influence the proportion between the number of clusters in areas where data elements are dense as opposed to their sparse regions. Finally, the algorithm—by the detection of oneelement clusters—allows identifying atypical elements, which enables their elimination or possible designation to bigger clusters, thus increasing the homogeneity of the data set.
The paper deals with the issue of reducing the dimension and size of a data set (random sample) for exploratory data analysis procedures. The concept of the algorithm investigated here is based on linear transformation to a space of a smaller dimension, while retaining as much as possible the same distances between particular elements. Elements of the transformation matrix are computed using the metaheuristics of parallel fast simulated annealing. Moreover, elimination of or a decrease in importance is performed on those data set elements which have undergone a significant change in location in relation to the others. The presented method can have universal application in a wide range of data exploration problems, offering flexible customization, possibility of use in a dynamic data environment, and comparable or better performance with regards to the principal component analysis. Its positive features were verified in detail for the domain’s fundamental tasks of clustering, classification and detection of atypical elements (outliers).
Szymon Łukasik, Konrad Lalik, Piotr Sarna, Piotr A. Kowalski, Małgorzata Charytanowicz and Piotr Kulczycki
Extracting useful information from astronomical observations represents one of the most challenging tasks of data exploration. This is largely due to the volume of the data acquired using advanced observational tools. While other challenges typical for the class of big data problems (like data variety) are also present, the size of datasets represents the most significant obstacle in visualization and subsequent analysis. This paper studies an efficient data condensation algorithm aimed at providing its compact representation. It is based on fast nearest neighbor calculation using tree structures and parallel processing. In addition to that, the possibility of using approximate identification of neighbors, to even further improve the algorithm time performance, is also evaluated. The properties of the proposed approach, both in terms of performance and condensation quality, are experimentally assessed on astronomical datasets related to the GAIA mission. It is concluded that the introduced technique might serve as a scalable method of alleviating the problem of the dataset size.