CSI 991

R Graphics

Fridays 3:00pm -- 5:00pm, Innovation Hall, Room 203

CSI 991 Section X01

Contacts:
csutton@gmu.edu
jgentle@gmu.edu

Some links on this page are available only to members of the seminar group.

Members Enter

The seminar will be conducted in the form a workshop. Participants are encouraged to put R on their own laptops, and to bring them to the seminars.

The main reference text is R Graphics, by Paul Murrell. We will work through the examples in the text, approximately one chapter per week, beginning with Chapter 2. The code for the graphs in the text are available at the author's website.
We want to understand everything about the example graphs, so hopefully someone will have figured out all of the details before the class presentation.

We will also discuss methods of incorporating graphs in TeX documents, and the R code necessary to produce other interesting graphs, such as those in The Elements of Statistical Learning, Data Mining, Inference and Prediction by Hastie, Tibshirani, and Friedman.

There are a number of useful books on R graphics and R in general. A list of books is available at the "Books" link on the main webpage for the R Project.

Frank Harrell has a very useful website on S/R resources. One of the links at that site is to a very useful introductory manual on S (and R) by Carlos Alzola and Harrell.

Other Things

Students should consider participating in the 2006 UCSD Student Data Mining Competition.

We will occasionally discuss various issues in statistical learning. There is an interesting article in the February, 2006, issue of Statistical Science (published by IMS, and available online to members), Classifier technology and the illusion of progress, by David Hand, with discussion.

We will also work on various exercises using R graphics. Exercises can be suggested by anyone, and all participants are encouraged to submit exercises.

Schedule

• June 9

  Installation of R.

  Packages in R. Which are most useful?
  • MASS
  • graphics (This is required for most of the plots in the text; see list on page 293.)
  • grid (See list on page 294.)
  • lattice (See list on page 296.)
  • maps
  • Hmisc
  • scatterplot3d
  • rgl
  
  

  Display of mathematical objects.

  Graphical displays are usually combinations of mathematical objects (plots) and text, which may contain simpler kinds of mathematical objects: symbols, variables, constants, etc.

  An important characteristic of a graph is the aspect ratio. The function eqscplot() in MASS helps to control this.

  The symbols for mathematical objects should be used consistently in a document. This means that labels in graphical displays should use fonts similar to those used in the text of the document. In R graphics this may require expression() and italic(). Other R expression tokens are similar to TeX without an escape character, e.g., sqrt(), hat(), theta, Theta, etc. Some tokens are available only in one font family, e.g. theta and Theta are always upright.

  Use "help(plotmath)" to see a description of the characters.
  Use "demo(plotmath)" to see a demo.

  The general syntax for producing text follows that of C; for example, "\n" gives a new line, "\"" gives a double quote mark, etc. Some of the old C escape sequences are irrelevant, such as "\?", and some don't work, such as "\b".

  Output file formats and methods of incorporation of graphs in documents.

  R draws displays on a "device". There are several devices and functions to control the devices (pp 20, 21).

  Some functions are platform-dependent, such as savefile() which is available in MS Windows.

  Traditional graphics in R; Chapter 2 of Murrell.

  Presentation by Jim Shine.

  ---

• June 16

  We will begin by clearing up several issues from Chapter 2, such as the orientation of the axes labels, the placement of plots in mfrow(), and even some simple things, like what is the bannerplot, and what is going on with barplot() in the examples.

  We will then proceed into Chapter 3, where the placement of plots will be considered more systematically.
  Presentation by Johan Bjursell.

  ---

• June 23

  Continuation of material in Chapter 3.
  Presentation by Ewa Gradzka.

  ---

• June 30

  Continuation of material in Chapter 3.
  Presentation by Keith Roberts.

  ---

• July 7

  3-D graphics

  Grid graphics: Chapter 4.
  Presentation by Johan Bjursell.

  ---

• July 11

  Continuation of material in Chapter 4.
  

  ---

Exercises

• Aspect ratio
  Generate 200 random numbers bivariate normal random variables with a correlation coefficient of 0.5. Compute the two principal components and plot the principal components on a scatterplot of the data. The PCs should be orthogonal.

• A new type of plot
  Develop an R function to produce the factorial model plot described in notes/ideas.
  • Extend this function to produce a "GLD" graph described in "Natural Induction and Conceptual Clustering", by Michalski et al., and illustrated in Figures 3 and 19 of that paper.
• 3-D graphics
  • Consider the conjugate gradient (CG) method for solving the linear system Ax=b. This method finds the minimum point on the quadratic surface x^TAx-2x^Tb. If A is 2-by-2, the surface is in three dimensions. Let A=matrix(c(5,2,2,3),nrow=2) and let b=c(7,5). The solution clearly is x=c(1,1).
    Show the progress of CG solution beginning with x⁽⁰⁾=c(4,-2). Use both contour and persp. Which is better?
  • Consider the nonlinear least squares problem BoxBOD in the NIST StRD collection.
    Using both contour and persp, plot the sum-of-squares surface. What salient features do you notice?
  • Compare both the usefulness and the user interfaces for contour and persp. Which is generally more useful? Which is easier to use?