Scribd
Upload
34 views2 pages

Exercise Sheet 5

The document discusses using principal component analysis (PCA) to visualize multivariate data through a series of exercises. PCA is applied to simulated bivariate and multivariate normal data to understand how it rotates and transforms the data. Biplots of real height and weight data are also examined to understand how to interpret distances and correlations from the plots.

Uploaded by

Swik
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
34 views2 pages

Exercise Sheet 5

The document discusses using principal component analysis (PCA) to visualize multivariate data through a series of exercises. PCA is applied to simulated bivariate and multivariate normal data to understand how it rotates and transforms the data. Biplots of real height and weight data are also examined to understand how to interpret distances and correlations from the plots.

Uploaded by

Swik
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

D. Metzler, M.

Hutzenthaler SS 2010
M ULTIVAR S TATS IN E COL AND G ENETICS — E XERCISES , SHEET 5

Principal component analysis (PCA) can be used to visualize multivariate data. We start with
self-generated data to get a feeling for the R command prcomp().

Exercise 1 Let’s see how PCA rotates 2-dimensional data.

• Sample 500 values from a multivariate normal distribution mean vector (0, 0)T and with
covariance matrix  
1 0
Σ= (1)
0 5
For this load the library mvtnorm and sample with the R command rmvnorm() as in the
lecture.

• Plot the data cloud with plot( ,xlim=c(-7,7),ylim=c(-7,7) ) (The cloud looks
roundish if the ranges are not fixed). Guess how PCA is going to rotate the cloud.

• Apply prcomp(,scale=FALSE) to your sampled data and store the returned object as
mypca. Have a look on the object mypca with unclass(mypca). Add the transformed
data cloud to the plot of the original cloud with points(mypca$x,col="red". Con-
sider the rotation matrix. Rounding its entries, which matrix does the rotation matrix resem-
ble?

• Have a look on the distance biplot and on the correlation biplot.

Exercise 2 Let us see how 5-dimensional data is visualized. In order to know the result in
R
advance, we start with 2-dimensional data and transform it into 5 .

• Sample 500 values from a multivariate normal distribution mean vector (0, 0)T and with
covariance matrix  
5 0
Σ= (2)
0 1
Denote the data matrix as x. Look at the data cloud (set the x-range and the y-range suitably).

• Multiply (matrix multiplication %*%) x with the matrix


 
0.5 0√ 0.5 0.5 0.5
(3)
0 −1/ 2 0.5 −0.5 0

• Now apply PCA. Try to plot the transformed data into the same plot as x such that the points
match. You might have to rotate or flip the transformed data (e.g. multiply the x-coordinate
with −1). For this note that the transformed data is 5-dimensional and that you only need
the first two columns.

• Have a look on the distance biplot and on the correlation biplot.


Exercise 3 This exercise trains you in reading biplots. Consider the correlation biplot and the
distance biplot of the height and weight data of the lecture.

• First execute the commands from the script to produce the biplots yourself. You find the file
HeightShoeWeight.txt on the course homepage.

• In the distance biplot ’Gewicht’ and ’Groesse’ are orthogonal to each other. Does that mean
that they are uncorrelated? You might want to confirm your answer with the cor() com-
mand.

• From the two biplots read off the approximate (Eucledian) distance between data point ’211’
and data point ’103’.

• Is the data point ’211’ a student with normal weight, overweight or underweight?

• Guess the sex of student ’121’.

Exercise 4 Let us continue as in Exercise 3. Consider the two biplots of the height and weight
data of the lecture.

• By looking at the biplots, order the students ’104’, ’106’ and ’226’ by increasing weight.
Hint: You need to use projections.

• The number of students used in the pca was N = 226. Looking at the lines in the correlation
biplot how well (in %) is weight represented by the first two principal components.

• Due the projection in the plane of the first two principal components, the angle between lines
in the correlation biplot might not well represent the correlation. Use the command cor()
to find out which of the correlations between weight, shoe and height is represented
poorly.

• By looking at the biplots, order the students ’98’, ’103’ and ’106’ by increasing shoe sizes.

Exercise 5 Download the data set ’EWU.txt’ from the web page. The variables X1, X2, X3 and
X4 have been explained in the lecture.

• Apply PCA to the data set and produce the distance biplot and the correlation biplot.

• Sort the countries GB, D, I and IRL according to their public dept level in 1997. Which of
these countries have a debt level below the average?

• Apply all three rules from the lecture in order to decide which of the principal components
are important.

You might also like