XGobi(1)XGobi(1)NAMExgobi - interactive dynamic graphics program for data visualization
SYNOPSISxgobi [ X options ] [ -subset n ] [ -only n/N or a,n ] [ -mono ] [
-version ] [ -std mmx|msd|mmd ] [ -scatmat ] filename
DESCRIPTIONxgobi is an interactive dynamic graphics program for data visualization
in the X Window System. It is especially designed for the exploration
of multivariate data. Its basic plot is a scatterplot, and these are
some of the tools available for scatterplot display and manipulation:
o Cycling rapidly through two-variable scatter plots.
o Three-dimensional rotation, including trackball control, in spaces
spanned by three variables at a time.
o Grand tours and correlation tours: smooth randomized sequences of
two-dimensional projections in order to explore a higher-dimen‐
sional point cloud of multivariate data. Projections can be manu‐
ally controlled and optimized with projection pursuit.
o Brushing: the ability to change the color or plotting character of
a point or a group of points.
o Hiding groups: using checkboxes to specify which brushed groups
should be plotted, or included in calculations.
o Scaling: moving and reshaping the plot interactively.
o Identification: displaying a label next to a point in the plot win‐
dow.
o Linked views: Brushing, identification and touring are linked; that
is, actions in the window of one XGobi process are immediately
reflected in another XGobi window displaying the same data.
o Line editing: Any pair of points can be connected with a line seg‐
ment.
o Moving points: Points can be moved on the screen. In higher-
dimensional views, the motion is interpreted as taking place in
some plane parallel to the screen.
o Smooths: Smooth curves can be drawn through point clouds.
o Subsetting: Subsamples can be drawn either systematically or ran‐
domly.
o Jittering: Variables can be jittered, which is useful for discrete
variables that take on only few values.
o Parallel coordinate display: shown in a separate linked window,
linked to the main window.
o Case label list: shown in a separate window, linked to the main
window for labeling points.
o Missing values: are accepted and can be dealt with by imputation of
constant values, random values, or user-supplied imputed values.
Missing value patterns can be examined in a separate linked XGobi
window.
o Variable transformations: A menu of transformations such as logs
and a few powers is available.
o Postscript plotting: The contents of the plot windows can be writ‐
ten out to postscript files for high-quality plotting.
o Online help: Most of the XGobi buttons and windows have associated
help files that can be displayed during an XGobi session. Click on
the Info button toward the top right of the window for instruc‐
tions.
XGobi has a direct manipulation interface, and all the above actions
are performed using the mouse.
XGobi can be used in conjunction with the S language for scientific
computing and data analysis. Execute help(xgobi) within S or Splus for
information; if the xgobi function has not been installed, investigate
the XGobi distribution files.
XGobi can also be programmed to use RPC (Remote Procedure Calls) to
communicate with other software.
OPTIONS-subset n
Specify the size n of the random sample of the data to be displayed
on startup. Although only the random sample is shown, all the data
is read in and available during the session. Use the subset panel in
Tools to change the sample size and sampling method during the ses‐
sion.
-only n/N
Specify the size n of the random sample of the data to be read in on
startup from a file of N rows. In constrast to subset, this option
does not read in all data.
-only a,n
Specify the initial row a and the number of rows n of the consecutive
block in data to read in. In constrast to subset, this option does
not read in all data.
-std mmx|msd|mmd
By default, the data are scaled into the plotting window using the
minimum and maximum values of each variable or variable group, in
such a way that the midpoint of the variable is at the center of the
plotting window and no points fall outside the window. Instead, to
scale using mean and largest distance, specify -std msd; to scale
using the median and largest distance, specify -std mmd.
-dev x
If you have specified -std msd or -std mmd, then you can also specify
the number of standard deviations (or median absolute deviations)
from the mean (or median) to be contained within the plotting window,
using the argument -dev x, where x is a real number between 0 and
100. The default is 2.
-mono
Emulate a black-and-white display. This simplifies producing pic‐
tures for many publications.
-version
Print a version date, the last date that any file in the release was
changed.
-scatmat
Build a scatterplot matrix from the input file, and display it in the
plotting window. This matrix is linked by default to non-scatmat
xgobi windows.
X options
The standard X command line options can be used with XGobi. These
include -display machinename:0, used when running an X program on one
machine and displaying its output on another, and -title Title, where
Title is a string you want to appear in the window manager titlebar.
FILES
XGobi accepts standard input, but is most often used with files, partly
because of the additional plot control that can be achieved using a set
of files. The data input file should be an ASCII file with the data
matrix arranged in rows and columns; in ASCII, rows must be distin‐
guished by carriage returns, and columns can be separated by any amount
of white space. Missing values can be coded as ".", "NA" or "na".
(The input file can also be a binary file, which can be produced within
XGobi once the ASCII data has been read in.) XGobi accepts other input
about the display of the data from files as well. If the data is in a
file named
filename
or
filename.dat
(either of which must be an ASCII file), or
filename.bin
(the binary version of the data), then the other files are as fol‐
lows:
filename.row
filename.rowlab
filename.case
Row or case labels: a label for each row of the data matrix, which is
displayed in the identification mode. The file should contain one
label per line.
filename.col
filename.collab
filename.column
filename.var
Column or variable labels: a label for each column of the data
matrix, which becomes part of the XGobi variable selection panel.
The file should contain one label per line.
To add a short label to be used in the parallel coordinates plot, use
the vertical bar "|" as a field separator and add the short name, so
each line looks like this: fullname|shortname.
filename.colors
Brushing colors: a color for each point in the plot, representing a
row or case of the data. The file should contain one color per line.
(It is probably best if the colors correspond to the colors used in
brushing; see the later section on resources.)
filename.glyphs
Brushing glyphs: a glyph type for each point in the plot, represent‐
ing a row or case of the data. The file should contain one glyph
type per line. The glyph types are as follows:
1 through 5: Five sizes of '+'
6 through 10: Five sizes of 'X'
11 through 15: Five sizes of open rectangle
16 through 20: Five sizes of filled rectangle
21 through 25: Five sizes of open circle
26 through 30: Five sizes of filled circle
31: A single-pixel point
filename.erase
Erase: a column of 1s (to have a point hidden on startup) and 0s (to
have the point plotted). There should be one value per line and as
many lines as there are rows in the data.
filename.lines
Line segments: specifications for the pattern of line segments which
connect pairs of points. The file should contain two numbers per
line. The pair of numbers represents the row numbers of the two
points that should be connected.
filename.linecolors
Line colors: a color for each line in the .lines file. The file
should contain one color per line. (It is best if the colors corre‐
spond to the colors used in brushing; see the later section on
resources.)
filename.nlinkable
The number of rows to be linked for brushing and identification. By
default, nlinkable is equal to the number of rows in the data. This
feature can be used to link ordinary scatterplots with plots that
have some decorations requiring additional points, such as clustering
trees.
filename.vgroups
Variable groups: an integer for each column in the data. Each set of
columns that is represented by the same integer will grouped together
for scaling and transformation. The file is just one long line of
integers. For example, an input file with four columns could have a
.vgroups file containing the line 1 2 2 3. The second and third col‐
umns are then grouped together. The range of their plotting axes is
be the same, and if column 2 is transformed, column 3 is transformed
at the same time.
filename.rgroups
Row groups: an integer for each row in the data. Each set of rows
that is represented by the same integer will grouped together for
brushing. In addition, line groups will be derived from the row
groups: all lines between points in the same row group will be in
the same line group. The file is just one column, and the numbers
don't need to be in any sort of sequence.
filename.missing
A file identical in structure to filename.dat, where non-zero values
indicate positions with missing (or censored, or otherwise excep‐
tional) values. This file represents the pattern of missing values
in the data; it can be examined in a separate XGobi window by select‐
ing Launch missing data XGobi... from the Tools menu.
filename.imp
Multiple imputations of missing values: Each column should have a
full set of imputed values. The number of rows needs to be identical
to the number of non-zero values in filename.missing, or the number
of missing codes in filename.dat if filename.missing is not pro‐
vided.. The imputed values should be given in their order in the
data column by column. For example, if filename.dat looks like this:
10 NA 12 -3
98 0 10 0
77 3 NA -5
1 2 NA 10
NA NA 5 -8
0 0 10 12
(six cases, four variables, five missing values), then filename.imp
with two sets of imputed values could look like this:
54 37
3 2
4 1
11 10
13 11
If the second column is selected for imputation (Select Impute miss‐
ing values from the Tools menu), the full data matrix with imputa‐
tions looks like this:
10 2 12 -3
98 0 10 0
77 3 10 -5
1 2 11 10
37 1 5 -8
0 0 10 12
filename.resources
Resources: a set of datafile-specific XGobi resources, which specify
the size of the plotting window and some user-selection option set‐
tings. The file is in the format of a standard X resource file. It
can be directly edited so that other resources can be specified. See
the later section on resource files for more information.
All of the above files can be created outside of XGobi, using an editor
or other UNIX utilities, and several of them (glyphs and colors, line
segments and line color, resources) can be written out during an XGobi
session, in which case they represent the results of interactions per‐
formed during that session.
X11 RESOURCE FILES
X11 resource files are used to control size, color, layout, labels, and
various other windowing features. The XGobi resources you are most
likely to find useful set the sizes of the plotting window and the
variable selection panel, the colors available for brushing, the font
types and sizes, and the title appearing in the title bar of the X win‐
dow. Some examples follow:
In this example, the first 5 of the 10 brushing colors are specified.
These colors should be drawn from the set of colors available on your
workstation or X terminal, which you can probably determine using the X
command 'showrgb.'
*brushColor0: white
*brushColor1: red
*brushColor2: green
*brushColor3: skyblue
*brushColor4: yellow
Here, the sizes of the plotting window and variable selection panel are
set. All units are in pixels. This file was created interactively
during an XGobi session.
*XGobi*PlotWindow.height: 120
*XGobi*PlotWindow.width: 490
*XGobi*VarPanel.width: 390
These two resources allow the variable widgets to be reduced in size,
which is helpful when there are many variables in the data.
*XGobi*VarWindow.width: 36
*XGobi*VarLabel.width: 36
Here, the title and font have been adjusted for making a videotape. An
explanatory title and a large font are used.
*XGobi.title: XGobi: Experimental Laser Data
*XGobi.iconName: XGobi: Laser
*plotFont: -*-lucida-medium-r-normal-sans-20-*-*-*-*-*-*-*
Several logical variables controlling XGobi's behavior can be defined
in a resource file. These are also represented by buttons, most of
which are found on the Option menu accessed from the main panel.
Descriptions can be found in the help window for each button.
*showAxes: True
*showPoints: True
*showLines: True
*linkBrush: True
*linkLineBrush: True
*linkIdentify: True
*linkTour: True
*plotSquare: True
*jumpBrush: True
*reshapeBrush: True
*carryVars: True
*glyphType: 6
*glyphSize: 1
*defaultPrintCmd: lpr -Pps1
AUTHORS
Deborah Swayne (dfs@research.att.com)
Dianne Cook (dicook@iastate.edu)
Andreas Buja (andreas@research.att.com)
CONTACT
Deborah Swayne (dfs@research.att.com)
SEE ALSO
Sample data files are included in the XGobi distribution.
For papers and further pointers, check the following web page:
http://www.research.att.com/areas/stat/xgobi/
XGobi supports Remote Procedure Calls (RPCs) as a means for interpro‐
cess communication (IPC). For example, a link among XGobi, ArcView, and
XploRe is currently being supported. For more details, see:
http://www.galaxy.gmu.edu/~symanzik/AXX/
BUGS
Probably too numerous to mention, but you do the authors a favor if you
report the ones you notice.
30 September 1997 XGobi(1)
