difference between mean2 and mean

I have a large matrix A of floats between 0 and 1. When I try mean2(A) it is very different from mean(A(:)). What could be the reason for this?

[James Tursa]

According to the doc mean2(A) uses mean(A(:)). How different are your results? On the order of eps of the answers?

[Andrei Bobrov]

please example

>> a = rand(1000);
>> mean2(a)
ans =
    0.5003
>> mean(a(:))
ans =
    0.5003

Answer 1

[Walter Roberson]

mean() takes special care for unsigned integers and signed integers. mean2() blindly converts everything to double and totals that and divides by the number of elements. If you are using an integer data type or signed integer data type, the mean2() approach could involve unnecessary loss of precision.

[Walter Roberson]

I have now submitted a documentation bug report about this, as the mean2() implementation is not as documented.

Answer 2

[Marc Jakobi]

That's strange. According to your code, there shouldn't be a different output (if A is of type double). It may be a different output if I is of type single.

The main difference is that mean2 computes the mean of all elements while mean computes the mean of each column of the matrix. However, mean2 converts the input (A) to double precision and mean doesn't. So if A is single precision the result may vary slightly if the matrix is very large.