6
$\begingroup$

I am currently studying principal component analysis in statistics. I am presented with the following example:

$\mathbf{X} = [X_1 \ X_2]^T$ a two-dimensional random vector with mean $\mathbf{\mu} = [0, -1]^T$, covariance matrix $\Sigma$, and eigenvalue-eigenvector pairs $(\lambda_j, \mathbf{\eta}_j), j = 1, 2$ given by $$\Sigma = \begin{bmatrix} 2.4 & -0.5 \\ -0.5 & 1 \end{bmatrix}, (\lambda_1, \mathbf{\eta}_1) = \left( 2.56, \begin{bmatrix} 0.95 \\ -0.31 \end{bmatrix} \right), (\lambda_2, \mathbf{\eta}_2) = \left( 0.84, \begin{bmatrix} 0.31 \\ 0.95 \end{bmatrix} \right)$$ Rotate the coordinate system: the first eigenvector agrees with the $x$-axis and the second with the $y$-axis.
Calculate projections in the new coordinates, new values $W_1$, $W_2$: $$W_1 = 0.95X_1 − 0.31(X_2 + 1) \ \ \ \text{and} \ \ \ W_2 = 0.31X_1 + 0.95(X_2 + 1)$$

I have some confusion with this example. Where did these calculations $W_1 = 0.95X_1 − 0.31(X_2 + 1) \ \ \ \text{and} \ \ \ W_2 = 0.31X_1 + 0.95(X_2 + 1)$ come from? In particular, where did the $(X_2 + 1)$ come from, whereas $X_1$ was chosen for the first term? Also what is meant by "the first eigenvector agrees with the $x$-axis and the second with the $y$-axis"?

$\endgroup$

1 Answer 1

7
$\begingroup$

We are subtracting from the mean, hence shifting the mean as the new origin.

We are told the mean is $[0, -1]$.

Hence we first subtract $0$ from $X_1$ and we subtract $-1$ from $X_2$. Hence that is how we obtain $(X_1, X_2+1)$.

After that we project it to the eigenvector, $(0.95, -0.31)$, this is done by taking the inner product of $(X_1, X_2+1)$ and $(0.95, -0.31)$.

Similarly for the second eigenvector.

$\endgroup$
2
  • $\begingroup$ Thanks! Any idea what "the first eigenvector agrees with the $x$-axis and the second with the $y$-axis" means? $\endgroup$ Feb 15, 2021 at 4:41
  • $\begingroup$ it's just a convention. after we map it, we can plot the result. so we plot $W_2$ vs $W_1$. , where $W_1$ is on the horizontal axis and $W_2$ on the vertical axis. $\endgroup$ Feb 15, 2021 at 4:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.