Multiscatter User's Manual

Multiscatter is a tool for exploring multidimensional data sets through the use of scatterplots.

Invoking multiscatter

Normally:	`multiscatter <inputfile`
For learning:	`multiscatter -demo`

For normal use:

multiscatter <inputfile

where inputfile contains many lines of numbers, like this:

69 -42 -13 -2
73 -45 -10 -1
72 -42 -13 -1
73 -48 -9 -1
72 -46 -10 -1
72 -41 -13 -2
69 -32 -20 -2
58 -14 -26 -3
52 -1 -30 -4
72 -44 -12 -2
68 -28 -22 -2
68 -29 -21 -2
72 -47 -9 -1
69 -44 -11 -2
66 -27 -22 -3
72 -43 -12 -1

Each line represents a single point in n-dimensional space, n being 4 in the above example.

For learning how to use it:

multiscatter -demo

causes multiscatter to create its own set of 3-dimensional data, whose structure the user can attempt to elucidate by exploration.

The controls

After you have made one or more scatterplots, you will probably select one point of interest by clicking on it. You can then use these buttons:

Recruit 1 neighbor: Finds the unselected point closest to the point or points already selected, and selects it. If several points tie for "closest", they all get selected; hence the "approximate" tilde on the button.
Recruit aggressively: Finds the unselected point closest to the point or points already selected, and notes how far it is from the closest selected point. All points within that distance are then selected, and the selection process is repeated until no unselected points are within that distance of a selected point. Repeated clicking of this button is useful for gathering together points that are clustered in a multidimensional way that might not be obvious from the two-dimensional projections of scatterplots.
Undo last recruit: Sometimes the Recruit aggressively button goes too far. A single level of "undo" is provided to help with this problem.
Complement selection: Selected points become unselected, and unselected points become selected.
Unselect all: All points become unselected.
Print selected points: The selected points are printed to standard output.
Delete selected points: The selected points are deleted from the dataset. If the remaining points are substantially more localized, newly made plots will have a larger scale.