SD4ft-Miner Demonstration - Selection of the attributes


Selection of the attributes

This section continues in explanation of the creating new task part described in previous section. Here will be shown how to:

Entering cedents

By cedent are meant antecedent, succedent and condition. Special features of the SD4ft-Miner are First set and Second set (as can be seen from fig. 1) but they are entered the same way as all other cedents so they will be described together as cedents. Each cedent consists of one or more partial cedents. There is already created one partial cedent called antecedent, succedent, condition, first set or second set. It only depends on user if another partial cedent will be used. Sometimes it is useful to use more partial cedents especially if it is desirable to influence more the final hypothesis. Each partial cedent has as a parametr minimal and maximal length. This length determine the minimum and maximum number of attributes (for each partial cedent) that will be in final hypotheses.

For example: if we have these three attributes (age, height, weight) in one partial cedent and the minimum length is set 1 and maximum length is set 2 than there will be only these combinations of attributes in final hypotheses:

Figure 1: Task window

Figure 1: Task window

Meaning of first set and second set

As was written above, SD4ft-Miner is similar to 4ft-Miner in the way that it seeks for association rules of the type:

Antecedent Succedent / Condition

But in the case of SD4ft-Miner is this rule applied on two different sets and so the previous formula is of the form:

Antecedent Succedent / Condition First set

Antecedent Succedent / Condition Second set

Association rules are computed for both sets and than are these computed values compared through SD4ft quantifiers such as difference of founded implication, difference of equivalence etc.

Selecting attributes

When partial cedents are defined it is possible to fill them up with attributes. For each attribute it is possible to set some parameters as can be seen from fig. 2.

Literal type is the first parameter and it has just two possibilities. Basic or Remaining. This parametr has reason when there are more then one attribut in partial cedent and can be combinated with length of the partial cedent for better results. Purpose of this parameter is, that at least one attribute set as basic will be in final hypothesis. Not every attribute set as basic but at least one.

Another parameter is Gace type. Gace type has three possibilities as can be seen from fig. 2 (positive, negative, both). Both means positive and negative. If positive gace type is selected then the attribute will be in final hypothesis in its positive form. If negative is selected then will be negation of this attribute in the final hypothesis.

The last parameter is Coefficient type. Coefficient types will be described as an example, so we suppose that these coefficients of attribute are given: 1, 2, 3 and the length of coefficient is set: minimum length = 1 and maximum length = 2:

in this example the subsets are {1}, {2}, {3}, {1, 2}, {1, 3}, {2, 3},
one category
here the maximum and minimum length is 1 so it is possible to choose only one of these coefficients {1}, {2}, {3},
is similar to subsets but when interval is used the coefficients must be one next to another. Intervals are in this example {1}, {2}, {3}, {1, 2}, {2, 3},
cyclic intervals
are similar to interval but here is possible to do the cycle from last coefficient to first. Cyclic intervals are {1}, {2}, {3}, {1, 2}, {2, 3}, {3, 1},
left cut
means the minimum and maximum length from the begining {1}, {1, 2},
right cut
means the minimum and maximum length from the end {3}, {3, 2},
means both left and right cuts {1}, {1, 2}, {3}, {3, 2},
boolean true
can be applied for boolean attributes and in this case only true values are considered,
boolean false
can be applied for boolean attributes and in this case only false values are considered,
both boolean
can be applied for boolean attributes and in this case both (true, false) values are considered.

Classes of equivalence

Classes of equivalence is useful tool when it is not desirable to have some group of attributes together in one hypothesis. Classes of equivalence will ensure, that there will be displayed only hypotheses where there will be only one attribut from the class of equivalence in the cedent.

Figure 2: Attribute's parameters

Figure 2: Attribute's parameters