The optimality indicator $f_i$ of the instance $xi$ is updated to $f^{'} i$ using the following formula:
$fi^{'}=f{i} + (\alpha{u}^{'} - \alpha{u}) y_{u} K(x_u,x_i) + (\alpha_l^{'} - \alphal ) y{l} K(x_l,x_i)$,
as shown in formula (13) in thundersvm-full.pdf, but at line 110 of the code smo_kernel.cpp, why the contribution of the two extreme instances $x_u$ and $x_l$ to $f_i$ is in the form of difference instead of summation, and in the update_f function, the update of $f_i$ is consistent with formula (13).
The optimality indicator $f_i$ of the instance $xi$ is updated to $f^{'} i$ using the following formula: $fi^{'}=f{i} + (\alpha{u}^{'} - \alpha{u}) y_{u} K(x_u,x_i) + (\alpha_l^{'} - \alphal ) y{l} K(x_l,x_i)$, as shown in formula (13) in thundersvm-full.pdf, but at line 110 of the code smo_kernel.cpp, why the contribution of the two extreme instances $x_u$ and $x_l$ to $f_i$ is in the form of difference instead of summation, and in the update_f function, the update of $f_i$ is consistent with formula (13).