BIC and ICL for 2000 and distinct numbers of components.

<p>The BIC figures reported are , where is the log likelihood of the model given the number of components, is the number of parameters in the model and is the sample size. The ICL is the same as the BIC plus a penalty term that increases if the predicted clusters overlap. The best estimate of the number of components is the one with the lowest value of the BIC or ICL.