Algorithm 3 Train HSVR

Input: $(x_i,y_i),i=0,\cdots n-1$ scales (output of Algorithm 1) 1: $ϵ=0.01({max}_i\left(y_i\right)-{min}_i\left(y_i\right))$ 2: $r_0=y=[y_0,\cdots ,y_{n-1}]$ 3: model = $\left[\right]$ # comment: empty list to hold the SVR model at each layer 4: $m=len\left(scales\right)$ # comment: number of HSVR layers 5: for i in range(0, m): 6:    ${\sigma }_i$ = scales[i] 7:    $C_i=5(max\left(r_i\right)-min\left(r_i\right))$ 8:    $sv{r}_i$ = fitted SVR on $(x,r_i)$ with parameters ${\sigma }_i$, $C_i$ and tolerance $ϵ$ 9:    predictions = $sv{r}_i$.predict(x) 10:    $r_{i+1}=r_i-predictions$ 11:    model.append($sv{r}_i$) 12: return: model