Algorithm 5 Pseudocode for multi:softprob in XGBoost

for each iteration do     for each training instance i do         Calculate predictions ${\widehat{y}}_i$ for all classes         Compute probabilities using the softmax function         for each class k do            Calculate $L(y_i,{\widehat{y}}_i)$ for class k         end for         Update the model to minimize the loss     end for end for