Base software environment

A considerable base of software was available to the students by using the Quail system (Oldford et al) which includes:

• the Common Lisp programming language including its powerful object system allowing multiple class inheritance and generic functions which can type on any number of arguments.
• a base programming environment supplied by the Macintosh Common Lisp vendor (Digitool, 1997) which is quite modern -- providing lisp file editors, incremental compilation, program steppers, process backtrace, structure inspectors, and some program analysis tools (who calls, etc.).
  Similar features are available for the PC from Franz Lisp for their Allegro Common Lisp (1997).
• Quail's multidimensional arrays including
  -- array mapping operators along any dimensions
  -- sorting, permuting, ranking along any dimensions
  -- usual matrix operators including solution of linear systems
  -- matrix decomposition objects (e.g. QR, LU, SVD, $\ldots$)
• Quail's statistical functionality including
  Summary statistics -- mean, median, percentiles, standard deviation, $\ldots$
  Data objects -- array objects containing some meta-data information.
  Model objects -- Extended Wilkinson-Rogers specification of generalized additive models (e.g. see Chambers and Hastie, 1992). Only the Gaussian linear model class is needed for the course. See Anglin and Oldford (1993) for further detail.
  Fit objects -- contains pointers to the model, the data, and results of fitting one to the other.
  Probability objects -- classes representing a standard suite of univariate discrete and continuous distributions with each providing a variety of probability calculations and pseudo-random variate generation.
• Quail's graphic objects. The model for these was first introduced in Hurley and Oldford (1991) and can be seen in an early demonstration in the video Hurley and Oldford (1988).
  Views and viewed-objects -- A graphic in Quail is a data structure called a view which can be displayed simultaneously in any number of viewports. The metaphor is that each graphic is a ``view'' of some other object, its viewed-object. Hence every view data structure retains a pointer to the viewed object.
  Compound views -- views which contain subviews. Compound views position their subviews in a display. The compound view and every subview may have its own viewed-object; subviews can themselves be compound views.
  Stock statistical graphics -- dotplots, boxplots, histograms, stem and leaf, 2 & 3D scatterplots, 2 & 3D function plots, scatterplot matrices, brushing, linking, $\ldots$
  Controls -- needle-sliders, bar-sliders, push-buttons, editable text-input, dialogs, pop-up menus. These could operate on anything.
  View layouts -- compound views which layout subviews in row, column, or grid fashion, or more generally at arbitrary positions specified by the user.
  Interactive display -- every view responds to three mouse buttons (left, middle, right) alone or in combination with two modifier keys (shift and ctrl). Unmodified mouse buttons typically produce menus which refer to the physical display of the selected view; ctrl-mouse buttons refer to the viewed-object of the selected view.
• Two strategic functions having methods for any object
  (Signposts   object $\ldots$) -- returns a list of ``signposts'' particular to the given object; each signpost is a kind of control button view which if displayed and mouse selected would lead to some other relevant display peculiar to that signpost from that object.
  (Display   object $\ldots$) -- returns a view, which if drawn would produce a reasonable display of the given object. Display always accepts a boolean argument :signposts? which if true will return a view augmented by signposts. Ctrl-middle-mouse on any view pops a menu offering the user the opportunity to call display on the viewed-object, with or without signposts. This means from a display, any viewed-object could be interacted with directly.

Other object oriented systems allowing user defined classes and generic functions could also be used. The critical base language features are the ability for the user to define classes having inheritance, generic functions, and to specialize system defined methods. Programmatic construction, display, and manipulation of statistical graphics is also essential.