please, could you explain how to compute this? Thanks

- In many applications, the classifier is allowed to "reject" a test example rather than classifying it into one of the classes. Consider, for example, a case in which the cost of a misclassification is $10 and the reward (negative cost) of correct prediction is $0.5. On the other hand, the cost of an additional human evaluation is only $3. We can summarize this by the following loss matrix Show that in general, for this loss matrix, but for any posterior distribution, there will be two thresholds 0 and 1 such that the optimal decision is to predict Y^=0 if p11 (where p1=p(Y=1x) ). - Compute 0 and 1. - Sometime, you can choose to double the consequence of your decision (for example when you are super-confident), which means your costs become 1 for the correct and 20 for the wrong prediction. Compute the new threshold 2,3 such that the optimal decision is to double the consequence of your prediction for Y^=1 when p1>2, or for Y^=0 when p1