Scoring metrices for multiclass classification

[ashwinshetgaonkar]

on page 201,
It is mentioned that to use micro average instead of the macro (default one) fol. code is written:
pre_scorer = make_scorer(score_func=precision_score,
pos_label=1,
greater_is_better=True,
average='micro')
how to use this in multiclass ? (pos_label=1 is fine for binary problems)

[rasbt]

It would be like shown in the equation on page 201. Say you have 3 classes, it would compute 3 TP (true positive) scores:

1. TP_1 by treating class 1 as the TP and class 2 & 3 as "rest"
2. TP_2 by treating class 2 as the TP and class 1 & 3 as "rest"
3. TP_3 by treating class 3 as the TP and class 2 & 1 as "rest"

Then, you do the same thing for the FP (false positives). Essentially, it works the same way as macro-averaging, but instead of having multiple precision scores that you average, you have multiple TP and the FP and compute the precision score from the combined individual scores.