Algorithm 8 Compute Adaptive Weights

Require:  $y,\widehat{y},focus\_pairs,focus\_multiplier$ $n\_classes\leftarrow columns\left(\widehat{y}\right)$ Initialize $class\_errors$ as a zero vector of length $n\_classes$ for $i=0$ to $n\_classes-1$ do     $class\_indices\leftarrow$ indices where $y=i$     $class\_errors\left[i\right]\leftarrow$ mean absolute error of $\widehat{y}[class\_indices,i]$ from 1 end for Normalize $class\_errors$ by its maximum value Initialize $weights$ as an empty dictionary for $i=0$ to $n\_classes-1$ do     for $j=i$ to $n\_classes-1$ do         if $i=j$ then            $weights\left[\right(i,j\left)\right]\leftarrow 0.1+0.1\times class\_errors\left[i\right]$         else            $avg\_error\leftarrow (class\_errors[i]+class\_errors[j\left]\right)/2$            $weights\left[\right(i,j\left)\right]\leftarrow weights\left[\right(j,i\left)\right]\leftarrow avg\_error$            if $(i,j)$ in $focus\_pairs$ or $(j,i)$ in $focus\_pairs$ then                $weights\left[\right(i,j\left)\right]\leftarrow weights\left[\right(j,i\left)\right]\leftarrow avg\_error\times focus\_multiplier$            end if         end if     end for end forreturn  $weights$