How many clusters?

A Consistent Markov Partition Process Generated from the Paintbox Process

“…See [1] for a relatively recent overview of this literature. Well-behaved mathematically tractable models of random partitions are of interest to probabilists as well as statisticians and scientists; see [10], [12], [13], and [15]. Ewens [10] first introduced the Ewens' sampling formula in the context of theoretical population biology.…”

Analysis and Learning Frameworks for Large-Scale Data Mining

“…We used 6 servers as slave machines for both of the proposed framework and Hadoop: 4 servers with 4-core 2.8 GHz CPU and 4 GB memory, and 2 servers with two of 4-core 2.53 GHz CPU and 2 GB memory. In Table 4 shows execution times of one iteration on three machine learning algorithms: K-Means [2], Dirichlet process clustering [12] and IPM perceptron [13,14]. The values are mean and standard deviation over 10 runs.…”

“…Table 4. Comparison of the parallel machine learning framework and Mahout on K-Means [2], Dirichlet process clustering [12] and IPM perceptron [13,14]. We also applied the framework in order to parallelize a learning algorithm of an acoustic model for speech recognition.…”

Clusterability assessment for Gaussian mixture models

“…Somewhat unfortunately, this heuristic method tends to fail in actual applications when the number of dimensions increases. More formal inspiration for clusterability analysis is provided by considerations on the number of clusters (main references include [3][4][5][6]) and component overlap analysis for mixtures of normal distributions (see [7][8][9][10]). However, both approaches originally assume the underlying partition to be already determined.…”