如何在MATLAB中计算MAPE和DS

Question

我使用fitrsvm进行默认验证、交叉验证和KFold验证。

%%In sample validation.
rng default ;
mdl = fitrsvm(X,Y, 'Standardize',true);
loss =  resubLoss(mdl)
%% out of sample validation with 80% traning and 20% validation    
CVmdl = crossval(mdl,'Holdout',0.2);
CVloss = kfoldLoss(CVmdl)
%% 10Fold Cross Validation Model
KFmdl = crossval(mdl);
KFloss = kfoldLoss(KFmdl)

我需要计算这些模型的MAPE和方向对称(DS)。在Matlab中是否有任何内置函数(如loss或KfoldLoss) ?还是我需要把它们作为函数来实现?

Answer 1

这里我实现了方向对称(DS)和平均绝对百分比误差(MAPE)作为函数

function mape( Y, Ypredict, indxtest)
smape = 0;
    if isempty(indxtest)
        for i = 1 :length(Y)
           smape = smape + (abs((Ypredict(i) - Y(i))) / Y(i));
        end
    else
        j = find(indxtest); 
        for i = 1 :length(j)
           smape = smape + (abs((Ypredict(i) - Y(j(i)))) / Y((j(i))));
        end   
    end
mape = smape * 100/length(Y)
end
-------------------------------------------------------
function ds( Y, Ypredict, indxtest)
sds = 0;
if isempty(indxtest)
   for i = 2 :length(Y)
      if (((Y(i)-Y(i-1))*(Ypredict(i)-Ypredict(i-1))) > 0)
         sds = sds + 1;
      end
   end
else
   j = find(indxtest);   
   for i = 2 :length(j)
       if (((Y(j(i))-Y(j(i-1)))*(Ypredict((i))-Ypredict(i-1))) > 0)
           sds = sds + 1;
       end
   end
end
ds = sds * 100/length(j)
end

虽然这对我来说很好，但如果有人帮助我改进它，要么减少行数，要么提高效率，我会很感激。