How to replace a number with 0 in an array

How to replace a number in a column with 0? I want to replace all 0.5 values in a column with 0

Answer 1

[dpb]

nCol=3;                 % say, for the particular column
x(x(:,nCol)==0.5)=0;    % turn to zero

NB: While 0.5 is exactly representable in floating point, exact comparison for floating point values is risky in general. See the FAQ <Why is 0.3-0.2-0.1 not equal to zero?>

For the general question look up "Using Logicals in Array Indexing" in the doc's; it's very important Matlab feature.

[Lilja Dahl]

Thanks alot. I am using data from a monitoring system and <0.5 stands for the detection limit of the instrument. It would also be nice to define <0.5 as random numbers from 0 to 0.5 (normal distribution).

What do you suggest? Thanks for your help

[dpb]

Previous comment where you asked this below:

"Just replace the RHS of the assignment with whatever is wanted; only restriction is you'll have to have saved the logical vector to know how many elements are needed to generate the proper number for the assignment. A constant is propagated automagically across any size target; multiple assignment on RHS must be conformant in size to the target on the LHS."

As for "numbers from 0 to 0.5 (normal distribution)", those two distributions don't really coincide. The normal is unbounded and you can only set a probability of not exceeding either range by adjusting the mean and variance from which you sample. If you take the crude way out and simply truncate samples outside the range, then you are, indeed, in the range but the result isn't normal.

I don't know what the measurement is but it doesn't seem likely to me that a RNV would be the most probable result of measurements taken with a more precise instrument of a higher sensitivity for the nondetectable measurements from this particular instrument.

It also seems to me that introducing such artificial data into a dataset is risky at best and potentially extremely misleading at worst.

Just because you can do something doesn't mean you should...

ADDENDUM

I still have strong reservations about the wisdom of doing this, but...if you were to really, really, really believe this is the right thing to do, I'd think the likely thing would be to use a half-normal with mean 0.5- and (say) 3-sigma at 0.01 or somesuch. Take abs() of the samples generated. This would put the largest number at just under the 0.5 cutoff value and spread them out towards zero probably roughly as you envsion. Truncating then for any that are <0 shouldn't pollute the resulting distribution too much. Try

hist(0.5-abs(1/6*randn(1,1000)))

to visualize the result. I'd still caution that there ought to be some way kept to keep track of which are "pseudo" values rather than real, though if you decide to do something of the sort.

Answer 2

[Jos (10584)]

A = [1 2 3 ; 0.5 0.5 2 ; 0.5 5 6] % example data
C = A(:,1)
tf = C == 0.5
A(tf,1) = 0

% or in one line: 
% A(A(:,1)==0.5,1) = 0

[Lilja Dahl]

I figured it out, thanks alot!

[Lilja Dahl]

and if I want to define A(tf,1) to be random numbers from 0-500 (normal distributed) instead of 0?

[dpb]

Just replace the RHS of the assignment with whatever is wanted; only restriction is you'll have to have saved the logical vector to known how many elements are needed to generate the proper number for the assignment. A constant is propagated automagically across any size target; multiple assignment on RHS must be conformant in size to the target on the LHS.

[vanam sindhuja]

Can I know the significance of tf.

[dpb]

...
tf = C == 0.5
A(tf,1) = 0
...

tf is the logical addressing vector; the variable name chosen to represent true|false as indicator it is a logical array.

As a subscripting expression, MATLAB returns only the locations in the referenced array for which elements in the indexing expression return TRUE