Compute accuracy, precision, recall, and F1 for single-label classification. For each class $c$, let $TP_c$ count correct predictions of $c$, $FP_c$ count predictions of $c$ whose true label differs, and $FN_c$ count occurrences of $c$ predicted as another class.

Use micro to aggregate class counts before computing metrics, macro to average class metrics equally, weighted to average them by true-label support, and binary to report the class selected by pos_label. A zero denominator contributes 0.0. Return accuracy, precision, recall, and f1 in a dictionary, with every value rounded to six decimals.