Binary and Multiclass Scores
Scores are used differently depending on the task.
In binary classification, the model may produce one score:
score = x . w + b
A positive score can mean class 1. A negative score can mean class 0. The threshold is often zero.
In multiclass classification, the model usually produces one score per class:
scores = [score_1, score_2, score_3]
The largest score is the model's current class choice before converting scores into probabilities.
scores = [1.2, -0.4, 2.1]
choice = class 3
Class 3 is chosen because 2.1 is the largest score.
A binary classifier uses threshold 0. It returns score -1.5. Enter 1 if the score is on the positive side, or 0 if it is on the negative side.
Compute it first, then check your number.
HintCompare with zero
Positive side means score greater than zero.
SolutionWork it out
The score -1.5 is less than 0, so it is on the negative side.
A model returns scores [0.7, 2.4, 1.9]. Using one-based class numbering, which class has the largest score?
Compute it first, then check your number.
HintFind the maximum
Compare the three raw scores.
SolutionWork it out
The largest score is 2.4, which is the second entry. With one-based class
numbering, the chosen class is 2.