How take Normalization for each column of 50x19 matrix?

I have the dataset like shown in the image below

Answer 1

[Adam Danz]

There are many interpretations and methods of normalization and the term is often used to describe standardization or rescaling so you'll need to specify what method you wish to use. Below I provide an example how to normalize columns of a matrix so that all values are between 0 and 1 according to the max and min within each column.

% fake data
data = randi(10, 6, 5); 

% Subtract min of each column 
datamin = data - min(data, [], 1); 

% Scale to the max of each column 
dataNorm = datamin ./ max(datamin, [], 1);

Now all values are of dataNorm are between 0 and 1 according to the max and min within each column. If this is what you meant by 'normalize', you can apply this to your data.

If you wanted to normalize via z-score, check out the normalize() function (from release 2018a). It also offers additional methods .

dataNorm = normalize(data, 1);

[Bajdar Nour]

Dear @Adam Danz I want to find the max. for each column and divide each element within the column by this max(i.e I have 19 cols. and I need 19 max., after that, I divide 50 elements in each row by its max.)

[Adam Danz]

You could use my example and make some small changes to achieve that. Using my fake data from above....

dataNorm = data ./ max(data, [], 1);

So you understand what's going on here, max(data, [], 1) gives you the max of each column (we specify the column dimension in the 3rd argument).

Then you must divide each element of data with the column-paired max by using " ./ ".

[Adam]

You may have to handle the case where a column is full of zeros though, if this is ever the case. You'll end up with Inf for your normalised values which may not be what you want (I assume you would want them to remain 0)

[Adam Danz]

Good point. One way to avoid that is to divide by 1 if the column contains all 0s.

columnMax = max(data, [], 1); 
columnMax(all(data==0,1)) = 1; 
dataNorm = data ./ columnMax;