Table 4.

Iris data: results for different $K$, under the standard prior (Sta) and the normal gamma prior (Ng), and when fitting an infinite mixture using the R package PReMiuM. The $MS{E}_{\mathit{\mu }}$ is calculated using the Mahalanobis distance based on Bayes estimates. Values given in parentheses refer to the draws associated with the number of non-empty components given in parenthesis in column ${\widehat{K}}_0$

Prior	$K$	${\widehat{e}}_0$	$e_0$ fixed	${\widehat{K}}_0$	$M_0$	$M_{0,\mathit{\rho }}$	$MCR$	$MS{E}_{\mathit{\mu }}$
Sta	3	0.34		3	10,000	0	0.027	0.338
	15	0.05		3 (4)	5,900 (4086)	0 (0.004)	0.027 (0.093)	0.336
	30	0.03		3 (4)	6889 (3111)	0 (0.002)	0.027 (0.093)	0.338
Ng	3		0.01	3	10,000	0	0.027	0.350
	15		0.01	3 (4)	7469 (2496)	0 (0.043)	0.033 (0.093)	0.343
	30		0.01	3 (4)	6157 (3730)	0 (0.147)	0.033 (0.093)	0.349
PReMiuM		$\widehat{\mathit{\alpha }}$		$K^{est}$			$MCR$	$MS{E}_{\mathit{\mu }}$
		0.52		2			0.33