Re: Patch for erfinv.m

[Alois Schloegl]

John W. Eaton wrote:
> On 24-Jan-2008, Alois Schloegl wrote:
>
> | The test is on x and the assignment in y.
>
> Doh, I was completely missing that detail.
>
> | So the problem is
> | x = [NaN, -2, -1, 0, 1, 2];
> | y = zeros(size(x));
> | y(abs (x) > 1) = NaN
> | 
> | y =
> |      0   NaN     0     0     0   NaN
> | 
> | The is no NaN in y for the case of isnan(x). 
> | 
> | The yields the (incorrect) result
> | erfinv(x)
> | ans =
> |      0   NaN  -Inf    -0   Inf   NaN
> | 
> | 
> | The correct result should be 
> | 
> | ans =
> |    NaN   NaN  -Inf    -0   Inf   NaN
>
> OK, then how about
>
>   ## x(i) < -1 or x(i) > 1 or isnan (x(i))  ==>  y(i) = NaN
>   y(abs (x) > 1 | isnan (x)) = NaN;
>   y(x == -1) = -Inf;
>   y(x == +1) = +Inf;
>   

Yes, this is working, too. That's also closer to the original test
 i = find ((x < -1) | (x > 1) | isnan(x));

Why do you prefer

  y(abs (x) > 1 | isnan (x)) = NaN;

to 

   y(~(abs(x) <= 1)) = NaN; 

?

Is it just the readability of the code, or do you expect some 
performance gain ?

Alois


> Please reply on the list.
>
> Thanks,
>
> jwe
>