Matlab 为什么这个高斯函数会给出不一致的参数误差？

输出1

function m=gaussian(med, var, n)
  if ( mod(n, 2)==0 )
      n=n+1;
  end;

  med=double(med);
  var=double(var);

  med = min (max(-(n+1)/2, med),  (n+1)/2);

  m=zeros(1,n);

  k1=(1/(2*pi*var)^0.5);
  k2=-0.5.*((med-(1:n)).^2)./var;

  m(1,1:n)=k1.*exp(k2);

输出2

>> gaussian([101 2 ; 3 4], [4 301 ; 2 1], [2 2])
error: gaussian: operator /: nonconformant arguments (op1 is 1x1, op2 is 2x2)
error: called from
    gaussian at line 13 column 5
>>

---

我不确定您希望得到的结果是什么，但您得到了这个错误，因为您正在通过

[2x2]

矩阵对标量（

[1x1]

维度）执行操作。注意，您正在进行矩阵除法（

/

运算符），而不是元素对元素的除法（

/

运算符）

但是，这并不是唯一的问题，因为以下行对于减号运算符也有类似的问题：

octave:1> function m=gaussian(med, var, n)
>   if ( mod(n, 2)==0 )
>       n=n+1;
>   end;
> 
>   med=double(med);
>   var=double(var);
> 
>   med = min (max(-(n+1)/2, med),  (n+1)/2);
> 
>   m=zeros(1,n);
> 
>   k1=(1/(2*pi*var)^0.5);
>   k2=-0.5.*((med-(1:n)).^2)./var;
> 
>   m(1,1:n)=k1.*exp(k2);
> endfunction
octave:2> debug_on_error (1)
octave:3> gaussian ([101 2 ; 3 4], [4 301 ; 2 1], 2)
error: gaussian: operator /: nonconformant arguments (op1 is 1x1, op2 is 2x2)
error: called from
    gaussian at line 13 column 5
stopped in gaussian at line 13
13:   k1=(1/(2*pi*var)^0.5);
debug> (2*pi*var)
ans =

     25.1327   1891.2388
     12.5664      6.2832

debug> 1/(2*pi*var) # matrix division
error: gaussian: operator /: nonconformant arguments (op1 is 1x1, op2 is 2x2)
error: called from
    gaussian at line 13 column 5
debug> 1./(2*pi*var) # your element by element division works
ans =

   0.03978874   0.00052875
   0.07957747   0.15915494

---

或者，可能函数没有错误，您得到这些错误是因为您错误地调用了函数。

非常确定该函数是为第三个参数（

n

）设计的，它是标量，而不是向量。@BenVoigt，scanar n给出了相同的错误。问题已编辑。输出将是一个向量，而不是一个矩阵。。。。所以它不会对联合分布进行抽样。可能这三个参数都必须是标量。看起来Matlab函数

mvnrd

可能就是您要寻找的。实际上，看起来这个函数是为单变量高斯函数编写的。

octave:1> function m=gaussian(med, var, n)
>   if ( mod(n, 2)==0 )
>       n=n+1;
>   end;
> 
>   med=double(med);
>   var=double(var);
> 
>   med = min (max(-(n+1)/2, med),  (n+1)/2);
> 
>   m=zeros(1,n);
> 
>   k1=(1/(2*pi*var)^0.5);
>   k2=-0.5.*((med-(1:n)).^2)./var;
> 
>   m(1,1:n)=k1.*exp(k2);
> endfunction
octave:2> debug_on_error (1)
octave:3> gaussian ([101 2 ; 3 4], [4 301 ; 2 1], 2)
error: gaussian: operator /: nonconformant arguments (op1 is 1x1, op2 is 2x2)
error: called from
    gaussian at line 13 column 5
stopped in gaussian at line 13
13:   k1=(1/(2*pi*var)^0.5);
debug> (2*pi*var)
ans =

     25.1327   1891.2388
     12.5664      6.2832

debug> 1/(2*pi*var) # matrix division
error: gaussian: operator /: nonconformant arguments (op1 is 1x1, op2 is 2x2)
error: called from
    gaussian at line 13 column 5
debug> 1./(2*pi*var) # your element by element division works
ans =

   0.03978874   0.00052875
   0.07957747   0.15915494

error: gaussian: operator -: nonconformant arguments (op1 is 2x2, op2 is 1x3)
error: called from
    gaussian at line 14 column 5
stopped in gaussian at line 14
14:   k2=-0.5.*((med-(1:n)).^2)./var;