Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Distributed k-Means streaming algorithms in Spark

Kyriakidou Ioanna

Full record

---

Summary

K-means is one of the most commonly used clustering algorithms that clusters the multi-dimensional data points into a predefined number of clusters. When data arrives in a stream, there is a need to estimate clusters dynamically, updating them on arrival. In this thesis, we will apply a sampling technique using a data structure called coreset trees, before any approximation algorithm is applied. Coresets are used to obtain a small weighted sample from the data stream. Using coresets in a tree-like form we successfully speed up the process of computing a summary of the original data. The advantage of such a coreset is that we can apply any clustering algorithm on a much smaller sample to compute a solution for the original dataset faster. In the second step, we are using a StreamKM++ to estimate the cluster centres of the summary. We evaluate the algorithm on how the parallelism level impacts the time needed to extract the clusters, finally we compare the consistency within clusters of data conclusions about the usage of coreset trees as a distributed sampling method.

Available Files

Services