Its weighted training error and scaling factor cm is computed (step 3b).The weights are increased for training samples, which have been misclassified (step 3c).