jgs ARFF § An ARFF (Attribute-Relation File Format) file is a text file that describes a list of instances sharing a set of attributes. § ARFF files were developed by the Machine Learning Project at the Department of Computer Science of The University of Waikato for use with the Weka machine learning software
jgs K-means Problems § K-Means clustering may cluster loosely related observations together. Every observation becomes a part of some cluster eventually, even if the observations are scattered far away in the vector space § Clusters depend on the mean value of cluster elements; each data point plays a role in forming the clusters. A slight change in data points might affect the clustering outcome. § Another challenge with k-means is that you need to specify the number of clusters (“k”) in order to use it. Much of the time, we won’t know what a reasonable k value is a priori.
jgs Algorithms § K-Means - distance between points. Minimize square-error criterion. § DBSCAN (Density-Based Spatial Clustering of Applications with Noise) - distance between nearest points. § Simple EM (Expectation Maximization) is finding likelihood of an observation belonging to a cluster(probability). Maximize log- likelihood criterion
jgs DBSCAN § The algorithm proceeds by arbitrarily picking up a point in the dataset § If there are at least N points within a radius of E to the point, then we consider all these points to be part of the same cluster. § Repeat until all points have been visited
jgs Simple EM (Expectation Maximization) EM can decide how many clusters to create by cross validation, or you may specify apriori how many clusters to generate. 1. The number of clusters is set to 1. 2. EM assigns a probability distribution to each instance which indicates the probability of it belonging to each of the clusters. 3. The training set is split randomly into 10 folds. 4. EM is performed 10 times using the 10 folds. 5. The loglikelihood is averaged over all 10 results. 6. If loglikelihood has increased the number of clusters is increased by 1 and the program continues at step 2. The number of folds is fixed to 10, if the number of instances in the training set is not smaller 10. If this is the case the number of folds is set equal to the number of instances.
jgs Evaluation § A clustering algorithm's quality can be estimated by using the logLikelihood measure, which measures how consistent the identified clusters are. § The dataset is split into multiple folds, and clustering is run with each fold. The motivation is that, if the clustering algorithm assigns a high probability to similar data that wasn't used to fit parameters, then it has probably done a good job of capturing the data structure. § Weka offers the CluterEvaluation class to estimate it