在matlab中给定两个矩阵计算一个混淆矩阵



我有两个Xval(预测值(和Sv(验证测试(矩阵,一个带有分类器输出数据,另一个带有相同样本的验证数据。每列表示预测值,例如[0 0 1 0 0 0 0 0]表示数字3(数字中的1(。我想知道是否可以用矢量化的方式或内置函数来计算混淆矩阵,两个矩阵的大小都是12000x10。生成两个矩阵的代码就是这个

load data;
load test;
[N, m] = size(X);
X = [ones(N, 1) X];
[Nt, mt] = size(Xt);
Xt = [ones(Nt, 1) Xt];
new_order = randperm(N);
X = X(new_order,: );
S = S(new_order,: );
part = 0.8;
Xtr = X(1: (part * N),: );
Xv = X((part * N + 1): N,: );
Str = S(1: (part * N),: );
Sv = S((part * N + 1): N,: );
v_c = [];
v_tx_acerto = [];
tx_acerto_max = 0;
c =  250;
w = (X'*X+c*eye(m+1))X' * S; 
Xval = Xv*w;
for i=1:12000
aux = Xval(i,:);
aux(aux == max(aux)) = 1;
aux(aux<1) = 0;
Xval(i,:) = aux;
end

存在内置函数confusionmat或plotcompusion。但如果你想完全控制,你可以自己写一个简单的函数,例如:

function [CMat_rel,CMat_abs] = ConfusionMatrix(Cprd,Cact)
Cprd_uq = unique(Cprd);
Cact_uq = unique(Cact);
NumPrd = length(Cprd_uq);
NumAct = length(Cact_uq);
% assert(NumPrd == NumAct)
% allocate memory
CMat_abs = NaN(NumPrd,NumAct);
CMat_rel = NaN(NumPrd,NumAct);
for j = 1:NumAct
lgAct = Cact == Cact_uq(j);
SumAct = sum(lgAct);
for i = 1:NumAct
lgPrd = Cprd == Cact_uq(i);
Num = sum( lgPrd(lgAct) == true );
CMat_abs(i,j) = Num;
CMat_rel(i,j) = Num/SumAct;
end
end
end

最新更新