Partitioning

  1. hfpsr generates a configuration file for a FIS. The number of fuzzy sets must be specified for each input, and for the output, as well as the hierarchy types to use for inputs and output. The created FIS has no rules.

    Java interface:

    Learning menu, Partitions submenu, Generate a FIS without rules option.

    Command line, hfpsr program:

    Arguments:

    Options:

    -oFIS' where FIS is the output configuration file name (default: 'data file name'-sr.fis)
    -oConj where Conj is the disjunction operator(default: prod)
    -vVertexFile where VertexFile is the vertex file (see hfpvertex, default: vertices are computed and a file is created).
    -f: sets the Classif='yes' output option

    Command line example:

    hfpsr iris '3 3 3 3' 1 0.01 3 3 MeanMax sum 0.01

    Creates the files iris.hfp, iris.vertex and iris-sr.fis with 3 fuzzy sets per input variable, the hfp hierarchy type, and a regular grid of 3 fuzzy sets for the output.

    or

    hfpsr iris '3 3 3 3' 2 0.01 3 3 MeanMax sum 0.01 -oiriskm.fis

    Creates the files iris.hfp, iris.vertex and iriskm.fis, where the input partitions are generated by the k-means procedure.

    hfpsr iris '2 2 3 3' 3 0 0 3 sugeno sum 0 -oirisreg.fis -f

    Creates the files iris.hfp, iris.vertex and irisreg.fis, with 2 MF regular grids for the fist two inputs, 3 MF regular grids for the other two, and a crisp classification output.

  2. Select partitions

    This method requires several programs.

    Java interface:

    1. Learning menu, Partitions submenu, HFP MF submenu, Generate a HFP file and Edit a HFP file options
    2. Learning menu, Partitions submenu HFP MF submenu, Generate vertices option.

      Vertices can be viewed with the View vertices option. The maximum number of vertices generated for each input is specified in the HFP (default value=7).

    3. Learning menu, Partitions submenu HFP MF submenu, Select partition option.

    Command lines:

    1. hfpcfg to create the HFP configuration file

      Arguments:

      • data file name
      • number of columns to ignore. Typically 1, which means to ignore the last column, representing the output.

      Two optional parameters:

      The ouput hfp file name (default value the data file name.hfp)
      The hierarchy type:1 for hfp, 2 for k-means, 3 for a regular grid (default value)

      Command line examples:

      hfpcfg iris 1

      Creates the file iris.hfp with a regular grid hierarchy

      or

      hfpcfg iris 1 iriskm.hfp 2

      Creates the file iriskm.hfp with a k-means hierarchy.

      The default generated output is crisp, with respective aggregation and defuzzification operators 'sum' and 'sugeno', and without the classification option.

      These parameters can be modified by editing the iris.hfp file. For the iris data, the classification option should be chosen, as the output is a variety.

    2. hfpvertex to calculate the fuzzy sets bounds in the various partitions

      Arguments:

      • data file name
      • HFP configuration file name

      Two optional parameters:

      -oVertexFile where VertexFile is the name of the output vertex file (default: vertices.'hierarchy type', i.e. regular, kmeans ou hfp)
      -kn This option only concerns the hfp hierarchy. If given, n is the number of groups used to form the initial partition in the call to the k-means algorithm. Otherwise, the initial partition is formed by grouping data into unique values, according to the tolerance given in the hfp configuration file.

      Command line example:

      hfpvertex iris iris.hfp

      Creates the file vertices.regular, if the hierarchy is a regular grid

      hfpvertex iris iriskm.hfp -oiris.vertices

      Creates the file iris.vertices for the kmeans hierarchy

    3. hfpselect to automatically select the number of MFs per variable.

      Arguments:

      • the FIS configuration file name
      • the data file name

      Optional parameters:

      -r: choose wm as rule induction method (default: fpa is used)
      -tx where x is the strategy to determine the data subset used to initialize the rule conclusion, 0 for MIN, 1 for DEC (default value).
      -my where y is the minimum matching degree (default value: 0.3)
      -ez where z is the minimum cardinality (default value: 3)
      -oFIS where FIS is the output FIS configuration filename (default value 'system name'.fis)
      -sw where w is the minimum cumulated weight for a rule to be generated (0.0 default value)
      -bc where c is the minimum coverage level(default value: 1.0, 100 %)
      -nf where f is the initial number of fuzzy sets per variable (default value: 1)
      -ig ehere g is the maximum number of iterations (default value: 10)
      -lFileV where FileV is the name of the vertex file, created by hfpvertex (default value: vertices.Hierarchy)
      -pNum where Num is the output number (défault: 0)
      -vFileTest where FileTest is the name of a data file used for validation (default value: the filename given in the second argument)

      Command line example:

      hfpselect iris iriskm.hfp -b0.7 -e2 -m0.3 -liris.sommets

      This program creates two files called result and result.min. The first one has as many lines as the number of attempts to complexify the FIS. In the second one, only the configurations kept at each step appear

      Reminder of the command lines for the partition selection procedure:

      - hfpcfg iris 1 iriskm.hfp 2

      - hfpvertex iris iriskm.hfp

      Set the output Classif flag to 'yes' in the iriskm.hfp file.

      - hfpselect iris iriskm.hfp -b0.7 -e2 -m0.3

      Note: As described in the user documentation, the four fields called In1, In2, In3, In4 indicate the number of MFs per input variable.