Written exercises

Do exercises 8-11 and 13 in the course notes on eigendecompositions.


The dataset heads in the course directory consists of six measurements on the heads of 200 soldiers in the Swiss army. The six variables are

  • MFB minimum frontal breadth
  • BAM breadth of angulus mandibulae
  • TFH true facial height
  • LGAN length from glabella to apex nasi
  • LTN length from tragion to nasion
  • LTG length from tragion to gnathion

These data were obtained from the R package Flury. Perform a PCA on these data. Specifically:

  1. Obtain the eigendecomposition of the sample covariance matrix, and interpret the first and second eigenvectors.
  2. Find the correlation between the six original variables and the first two principal components (a six by two matrix). Make a scatterplot of these correlations, draw a unit circle on the plots, and interpret the plot.
  3. For $q=0,\ldots,6$, how much variation in the data is explained by the best $q$-dimensional affine approximation?
  4. The Swiss army is considering ordering gas masks, and are debating between making either two types of gas masks or four. If they make two types, describe how the two types should differ from one another. If they make four types, describe what the differences should be. Do you think they should make two or four types?


The phoneme data contains $n=4509$ periodograms, each periodogram consisting of $p=256$ values and representing an audio sample of someone speaking a sound. More information on these data can be found here.

  1. Obtain the eigenvalues of the sample covariance matrix. Make a plot of the ordered eigenvalues and the cumulative proportion of variance explained. What fraction of the variance in the data is explained by the first four principal components?
  2. The rownames of the data matrix indicate the sounds that each periodogram represents. Make scatterplots of the first four principal components, indicating the different sounds by plotting color, character or text.
  3. Compute the four-dimensional vector of principal component means for each of the five different sounds. Report these values and include them on the previous plots.
  4. Categorize each row of the data matrix to the sound category that it is closest to, in terms of the first four principal component scores. Make a five-by-five two-way table indicating how the different sounds are categorized.