floor((I3(​i,j-1)+I3(​i-1,j))/2)

the expression evaluates to 192.5, but the function returns 193 instead of 192. Please help me out with this problem.

[Fangjun Jiang]

What do you get for this?

floor(192.5)

ans = 192

[RENJI]

In my program it is 193, but when I type in command prompt it is 192.

[RENJI]

I want it in my computation, not just for printing the value...in the whole computation there is a mismatch because of this rounding error.

Answer 1

[Bruno Luong]

@RENJI "the expression evaluates to 192.5"

The above is a FALSE statement since (I3(i,j-1)+I3(i-1,j))/2 cannot have fractional digit provided that I3 is of class uint32.

Solution: convert your array to double to do floating point arithmetics

I3d = double(I3);
floor((I3d(i,j-1)+I3d(i-1,j))/2)

[RENJI]

Thank you very much...its working fine now...thanks for your constant support.

Answer 2

[Sulaymon Eshkabilov]

To get the displayed data w.r.t your format specs:

ANS = 192.5;
% One decimal point
fprintf('Solution =  %.1f \n', ANS)

Solution =  192.5 

% OR two decimal points
fprintf('Solution =  %.2f \n', ANS)

Solution =  192.50 

% OR round up towards to -inf as Fangjun suggested 
fprintf('Solution = %d \n', floor(ANS))

Solution = 192 

Answer 3

[Bruno Luong]

Set a breat point in your program, when it breaks type this and report the result

I3(i,j-1)
I3(i-1,j)
(I3(i,j-1)+I3(i-1,j))/2
floor(ans)

exist('floor')
which floor((I3(i,j-1)+I3(i-1,j))/2)

Answer 4

[Walter Roberson]

You values are integer data class, such as uint8() . When you do calculations on integer data type, the result is calculated as if the values were converted to double, then the value was calculated, and then the result was cast to the integer data type. But when you cast a floating point number to an integer data type, the value is rounded. So the internal 192.5 that is calculated from the / operation is uint8() and converting to integer class rounds so it is rounded to uint8(193) . Then you are applying floor() to the uint8 193 which of course is still 193.

[RENJI]

how to resolve it?

[RENJI]

I3 array is int32.

[Walter Roberson]

floor((double(I3(i,j-1))+double(I3(i-1,j)))/2)

but if you were doing a bunch of these, convert the entire I3 array to double and index the double array.

D3 = double(I3);
floor((D3(i,j-1)+D3(i-1,j))/2)

[RENJI]

Thank you very much...its working fine now...