CODEWITHSUNDEEP
matlabmachine-learningnormalizationscaling
Khurram Majeed
Khurram Majeed

Reputation: 2411

Z-scores using specified mu and sigma

I have converted my training data matrix into z-scores for each column. I have mu and sigma for each column from the output of zscore.

I also have another matrix (my test data) and I want to convert it into z-scores using the mu and sigma obtained in previous the step. My implementation uses for loops as shown below:

TEST_DATA = zeros(num_rows,num_cols,'double');

for rowIdx = 1:num_rows,
    for colIdx = 1:num_cols,
        TEST_DATA(rowIdx,colIdx)=(input(rowIdx,colIdx)-MU(colIdx))/SIGMA(colIdx);
    end
end

Is there any faster way of achieving this in MATLAB?

Upvotes: 1

Views: 3133

Answers (2)

H.Muster
H.Muster

Reputation: 9317

You can use bsxfun:

%// Sample data
matrix = rand(10, 10);
testData = rand(10, 10); 

%// Obtain mu and sigma
mu = mean(matrix, 1);
sigma = std(matrix, [], 1);
%// or use: [Z, mu, sigma] = zscore(matrix);

%// Convert into z-scores using precalculated mu and sigma
C = bsxfun(@rdivide, bsxfun(@minus, testData, mu), sigma);

Upvotes: 2

Peter
Peter

Reputation: 14947

The documentation of zscore explains that it simply subtracts the mean and divides by the standard deviation. The only tricky part is apply the vectors of mu/sigma to each column. But if you don't know how to do the fancy way, do it with a for loop. I'll leave it this way for readability. If you need faster, look into bsxfun.

for ii=1:size(mat,1)
    mat(ii,:) = (mat(ii,:) - mu) ./ sigma;
end

Upvotes: 0

Related Questions