# “Normalizing” variables for SVD / PCA

Suppose we have $N$ measurable variables, $(a_1, a_2, \ldots, a_N)$, we do a number $M > N$ of measurements, and then wish to perform singular value decomposition on the results to find the axes of highest variance for the $M$ points in $N$-dimensional space. (Note: assume that the means of $a_i$ have already been subtracted, so $\langle a_i \rangle = 0$ for all $i$.)

Now suppose that one (or more) of the variables has significantly different characteristic magnitude than the rest. E.g. $a_1$ could have values in the range $10-100$ while the rest could be around $0.1-1$. This will skew the axis of highest variance towards $a_1$'s axis very much.

The difference in magnitudes might simply be because of an unfortunate choice of unit of measurement (if we're talking about physical data, e.g. kilometres vs metres), but actually the different variables might have totally different dimensions (e.g. weight vs volume), so there might not be any obvious way to choose "comparable" units for them.

Question: I would like to know if there exist any standard / common ways to normalize the data to avoid this problem. I am more interested in standard techniques that produce comparable magnitudes for $a_1 - a_N$ for this purpose rather than coming up with something new.

EDIT: One possibility is to normalize each variable by its standard deviation or something similar. However, the following issue appears then: let's interpret the data as a point cloud in $N$-dimensional space. This point cloud can be rotated, and this type of normalization will give different final results (after the SVD) depending on the rotation. (E.g. in the most extreme case imagine rotating the data precisely to align the principal axes with the main axes.)

I expect there won't be any rotation-invariant way to do this, but I'd appreciate if someone could point me to some discussion of this issue in the literature, especially regarding caveats in the interpretation of the results.

-

## migrated from math.stackexchange.comJun 22 '11 at 3:30

This question came from our site for people studying math at any level and professionals in related fields.

The problem itself usually is not rotation invariant, because each of the variables is recorded with a conventional unit of measurement appropriate to it. E.g, $a_1$ might be in feet, $a_2$ in microns, $a_3$ in liters, etc. Even when all units are the same, if the variables measure different kinds of things, the amounts by which they vary will likely differ in ways characteristic of those variables: once again, this is not rotation invariant. Therefore you should abandon rotation invariance as a guiding principle or consideration. –  whuber Jun 22 '11 at 19:28

A common technique before applying PCA is to subtract the mean from the samples. If you don't do it, the first eigenvector will be the mean. I'm not sure whether you have done it but let me talk about it. If we speak in MATLAB code: this is

clear, clf
clc
%% Let us draw a line
scale = 1;
x = scale .* (1:0.25:5);
y = 1/2*x + 1;

y = y + rand(size(y));

%% plot and see
subplot(1,2,1), plot(x, y, '*k')
axis equal

%% Put the data in columns and see what SVD gives
A = [x;y];
[U, S, V] = svd(A);

hold on
plot([mean(x)-U(1,1)*S(1,1) mean(x)+U(1,1)*S(1,1)], ...
[mean(y)-U(2,1)*S(1,1) mean(y)+U(2,1)*S(1,1)], ...
':k');
plot([mean(x)-U(1,2)*S(2,2) mean(x)+U(1,2)*S(2,2)], ...
[mean(y)-U(2,2)*S(2,2) mean(y)+U(2,2)*S(2,2)], ...
'-.k');
title('The left singular vectors found directly')

%% Now, subtract the mean and see its effect
A(1,:) = A(1,:) - mean(A(1,:));
A(2,:) = A(2,:) - mean(A(2,:));

[U, S, V] = svd(A);

subplot(1,2,2)
plot(x, y, '*k')
axis equal
hold on
plot([mean(x)-U(1,1)*S(1,1) mean(x)+U(1,1)*S(1,1)], ...
[mean(y)-U(2,1)*S(1,1) mean(y)+U(2,1)*S(1,1)], ...
':k');
plot([mean(x)-U(1,2)*S(2,2) mean(x)+U(1,2)*S(2,2)], ...
[mean(y)-U(2,2)*S(2,2) mean(y)+U(2,2)*S(2,2)], ...
'-.k');
title('The left singular vectors found after subtracting mean')


As can be seen from the figure, I think you should subtract the mean from the data if you want to analyze the (co)variance better. Then the values will not be between 10-100 and 0.1-1, but their mean will all be zero. The variances will be found as the eigenvalues (or square of the singular values ). The found eigenvectors are not affected by the scale of a dimension for the case when we subtract the mean as much as the case when we do not. For instance, I've tested and observed the following that tells subtracting the mean might matter for your case. So the problem may result not from the variance but from the translation difference.

% scale = 0.5, without subtracting mean
U =

-0.5504   -0.8349
-0.8349    0.5504

% scale = 0.5, with subtracting mean
U =

-0.8311   -0.5561
-0.5561    0.8311

% scale = 1, without subtracting mean
U =

-0.7327   -0.6806
-0.6806    0.7327

% scale = 1, with subtracting mean
U =

-0.8464   -0.5325
-0.5325    0.8464

% scale = 100, without subtracting mean
U =

-0.8930   -0.4501
-0.4501    0.8930

% scale = 100, with subtracting mean
U =

-0.8943   -0.4474
-0.4474    0.8943


-
I should have mentioned in the question that the mean has already been subtracted. I'll edit it accordingly. –  Szabolcs Jun 22 '11 at 7:37
One might simply divide each variable by its standard deviation, but I was wondering if there are other things people do. For example, we can think of this dataset as a point cloud in $N$-dimensional space. Is there a way to do it in a way that does not depend on the rotation in this $N$-d space? If we divide by standard deviations, it will matter along which axes those standard deviations are taken (i.e. it's not rotation invariant). If we do it along the principal axes, then I think the variables will appear uncorrelated. –  Szabolcs Jun 22 '11 at 7:41
I realize there might not be a rotation-invariant way to do it, but I'd love to at least read some discussion of these issues ... any pointers welcome. Note: I have no training in applied stat (only maths, such as linalg, prob theory), so I'm learning this stuff as I'm going. –  Szabolcs Jun 22 '11 at 7:43
When you don't center the samples (subtract the means from the columns), the first eigenvector usually is not the vector of means. –  whuber Jul 9 at 20:46

The three common normalizations are centering, scaling, and standardizing. With variable X:

Centering is Xi-MEANx. The resultant X will have mean=0.

Scaling is Xi/sqrt(SSx). The resultant X will have SS=1.

Standardizing is centering-then-scaling. The resultant X will have mean=0 and SS=1.

-
Can you define "SS" please? –  Szabolcs Jun 22 '11 at 7:53
Sum-of-squares. Sum of squared Xi. –  ttnphns Jun 22 '11 at 7:57
The reason for setting the sum of squares to 1, and not the variance, is that then the singular values will correspond to the standard deviations along the principal axes (unless I'm mistaken)? –  Szabolcs Jun 22 '11 at 8:08
Please also see my edit to the question. –  Szabolcs Jun 22 '11 at 8:13
@Szabolcs, I actually may miss a point of your edit. But PCA (or SVD) is just a rotation itself (a special case of orthogonal rotation of the axes). Any translation (like centering) or shrinking/dilatation (like scaling) of the cloud should affect the results of this rotation. –  ttnphns Jun 22 '11 at 8:40

To normalizing the data for PCA, following formula also used

$\text{SC}=100\frac{X-\min(X)}{\max(X)-\min(X)}$

where $X$ is the raw value for that indicator for country $c$ in year $t$, and $X$ describes all raw values across all countries for that indicator across all years.

-
Who said anything about countries or years? –  Nick Stauner Jul 9 at 11:21