Inference Window

There are two ways to infer using FisPro: inference using a data file, which is available from the Data menu, and will be described later, and graphical inference, which we now detail. Graphical inference is displayed as an array. Each line corresponds to an active rule, and each column to one fuzzy set in the rule premise or conclusion.

The user can change the value of an input variable by moving a cursor, or entering the numerical value, and see the effect onto the fuzzy inference system output. The matching degree of each rule is displayed, as well as the result of rule aggregation.

The inference window is for educational purposes mainly and is limited to the use of small size systems, in terms of variable and rule number.

This window is implemented as a table structure, this allows rule sorting using a click in any column header.

Any change in other windows takes effect immediately. The inference being done dynamically, the results displayed correspond to the current configuration.

Implicative rules

Two inference mechanisms are implemented in Fispro. The mechanism used by conjunctive rules is called FITA (First Infer Then Aggregate) and allows to separately infer a conclusion value for each rule before aggregating these values. The same mechanism is also used for implicative rules when input data are precise values. If this is not the case, when at least one input value is imprecise, a different mechanism is requiered, called FATI (First Aggregate Then Infer).

For a detailed comparison of the two types of rules, the inference mechanisms and the FATI implementation chosen in Fispro, see [13].

In the interface, when the FATI algorithm is used, the rule conclusion corresponding to the input data is not displayed as it cannot be calculated.

Fuzzy data

The inference window menu includes an option called Fuzzy Data, with is again divided into two options:

• Template: used to transform each precise input into a fuzzy number centered on that value. Two parameters are required: the kernel and support width. The support width is constrained to be bigger than or equal to the kernel one. A different template can be defined for each input, it can be saved into a file with .tpl extension, and reloaded from such a file.

  In the inference itself, the cursor allows to move the central value, and the fuzzy number will be automatically built using this central value and the given template. Manual inference with fuzzy data is available for all types of rules. It is limited to 2 inputs and 1 output if the system has at least one implicative output.

  This option is not available yet for the inference from a data file.

• Alpha-cut: allows to specify the $\alpha$ cut number used to approximate the imprecise input value in the FATI inference mechanism. This number is common to all system inputs.