Parallel k-means Project Update
Busy times at my workplace at BMW - yet, there are some updates with regard to the parallel k-means project. Besides extending the functionality of the matrix and vector classes, I implemented a new k-means initialization method which is known as KKZ initialization. The main characteristic is its deterministic nature in contrast to the random approaches which have been already implemented. Using this initialization scheme, the k-means algorithm always results in the same cluster assignments - given that, for each run, the same dataset is applied. The reason for its inclusion in the parallel k-means library are the good experiences I gained throughout several projects I tackled during my PhD period. For downloading the new version, please visit the Project page.