ClassConfidence
ClassConfidence[probability]
gives the confidence of a probability vector as produced by a network's softmax output, i.e. the same input ClassDecoder takes, without collapsing it to a class label. Confidence is 1 minus the normalized entropy of the vector, so it is 1 for a fully certain (one-hot) vector and 0 for a maximally uncertain (uniform) vector.
QMRITools`
QMRITools`
ClassConfidence
ClassConfidence[probability]
gives the confidence of a probability vector as produced by a network's softmax output, i.e. the same input ClassDecoder takes, without collapsing it to a class label. Confidence is 1 minus the normalized entropy of the vector, so it is 1 for a fully certain (one-hot) vector and 0 for a maximally uncertain (uniform) vector.