The User Interface

The user interface for RCLASS is started by typing RCLASS & at the UNIX command prompt or by selecting ``RCLASS'' in the main menu. Afterwards, the window in Figure  is shown.

  
Figure: Main window of RCLASS.

Files The program takes the name of three input files: table file, dictionary file and rule file. The first two are the same as those given to RGEN, and the rule file should be the output from RGEN. The result of the program is written to the output file, or standard out if no such is given. The output is equal to the input table file with a column add at the end, containing the classification of the system. An updated dictionary file (covering both the rule file and the new table file) is written with the name ``updated.dic''.
Method If there are conflicting rules for some of the objects, RCLASS uses the following methods to classify:

• First rule Use the first rule in the rule file which is applicable to the object.
• Best rule Use the rule with the highest accuracy which fits the object. In case of several rules with same accuracy, the first one is used.
• Linear voting Take all applicable rules into account. A rule gives a vote for its decision equal to the accuracy. The decision with the highest sum from all rules is used.
• Squared voting Same as linear voting except that the accuracy is squared for each rule.
• Exponential voting Same as linear voting except that an exponential function is applied to the accuracy and the result is used for voting.
• Weight of evidence The method of weight of evidence measure is used for voting.
  

Selecting Rule Base When RGEN traverses its nodes, see Chapter , and creates the rules, the nodes are numbered in the order they are visited. Each node take a specific set of attributes into account when generating rules. RCLASS allows the user to choose which nodes in the lattice that are of interest, and this is done by using the window shown in Figure , which is displayed when ``Select rulebase no.'' is chosen. The nodes are listed in the order they were traversed, and their attribute numbers, starting from 0, is shown to the right of the check button.

  
Figure: The rulebase selection window.

Selecting Attributes In addition to the ``Select rulebase no. '' option, an option called ``Select attributes'' is available, but is not possible to use both these options at once. ``Select attributes'' lets you select which attributes may be taken into account, and which attributes must be taken into account when classifying. This corresponds to a sub lattice containing some of the nodes, and only rules created at these nodes will be used for classification.

  
Figure: The attribute selection window.

The example in Figure  uses three attributes, numbered 0, 1 and 2. It is specified that the rules to use should come from nodes in the lattice which has no other attributes than that the attributes (0 1 2), but contains at least the attribute 0. The lattice for this example is shown in Figure . Rules from the four nodes (0 1 2), (0 1), (0 2) and (0) will be used for classification in this example. The thick lines and extra circles shown nodes in the sub-lattice from (0 1 2) to (0).

  
Figure: A lattice for three attributes.

Test Set this option if the data set contains a decision as the last column. RCLASS will then check its own decision against the correct decision, and create statistics. The output will contain RCLASS' decision, and not the original one. Some statistics, like the number of correctly classified objects, and the number of objects not recognized by any rule is printed.
Validity When ``validity'' is set, the program will add a final column to its output. This column will contain a number from 0 to 1000, and reflects the system's belief in its own decision, based on the accuracy for the rules used. The validity measure should not be taken literally, but rather as a hint to whether the classification is likely or not.
Full Format RCLASS prints its output by using the dictionary file to translate back to values from the real world if this option is set. Otherwise, the same values as in the table file are used.
Give Alternatives If ``give alternatives'' is set, the program is allowed to give several different possible values for the classification.
Ignore Blocks This option tells the program to ignore all blocks in the rule file.
Change Font By default the program uses a times roman font with size 12 points. This may be changed by setting one or two environment variables before starting the program:

RFONTSIZE The size of the font, in tenths. Defaults to 120.

RFONTDESCRIPTION A complete font description following the rules of the script language Tk. Defaults to *-times-medium-r-normal-*-120-*-*-*-*-*-*