Data mining — Rule filter constraints

You can limit the rules to be included in a model by applying rule filters in an Associations training run or in a Sequence Rules training run.

To filter rules, you can use the following constraints:

Count constraints
Item constraints
Range constraints

You can use these constraints in any combination within one single rule filter object.

Count constraints

If you have specified an upper limit for the number of rules to be included in a generated model, you can specify count constraints to select the 'best rules'. You can specify integers or strings.

The following table shows the valid key words and the corresponding integer IDs. The key words are sorted in alphabetical order.

Table 1. Valid values for count constraints
Count constraint	Integer	String
Accessible volume	28	accessibleVolume
Weight of the rule body	22	bodyWeight
Confidence	2	confidence
Weight of the rule head	21	headWeight
Weight of each item in rules or item sets	24	itemWeight
Lift	3	lift
Number of items	5	nbOfItems
Number of item sets	6	nbOfItemsets
Weight of the rule	20	ruleWeight
Support	1	support
Support*confidence	4	suppTimesConf
Support times weight of the rule	23	suppTimesRuleWeight
Elapsed time between adjacent parts of the rule	10	stepWindow
Total elapsed time from the beginning to the end of the rule	9	totalWindow
Average weight of the transactions (Associations) or transaction groups (Sequences) that support the rule	29	transactionWeight

Example of count constraints

This example specifies that the model must not contain more than 100 rules. If more rules are found, the 100 'best' rules are included in the model.

The best rules are selected according to the following criteria:

Number of items in rules: The best rules are the shortest rules. The shortest rules include the lowest number of items.
Largest lift value: If the previous criterion does not produce a clear ranking of the rules, for example, if there are 46 rules with 2 items and 2375 rules with 3 items, the rules that have the largest lift value are selected. This means that from the rules with 2 items and from the rules with 3 items, the rules with the largest lift value are selected as best rules.

IDMMX.DM_RuleFilter()   
   ..DM_setMaxNumRules(100)   
   ..DM_addCountConstr(1,'nbOfItems','ascending')   
   ..DM_addCountConstr(2,'lift','descending')

Item Constraints

With item constraints, you can limit the number of rules to be generated in a model. You can specify to include or to exclude items in the rule body, in the rule head, or in the complete rule.

If you want to include the specified items, the rules that contain at least one of the specified items are generated. Including specified items is called a positive item constraint.

If you want to exclude the specified items, the rules that contain any of the specified items are discarded from the results. Excluding specified items is called a negative item constraint.

You can combine positive item constraints and negative item constraints that apply to the rule body, the rule head, or to the complete rule by using the boolean predicates AND or OR. Item constraints that are connected by the predicate AND must be matched simultaneously. If you apply item constraints that are connected by the predicate OR, at least one of the item constraints must be satisfied.

Item constraints also work with the categories of a taxonomy. Taxonomies are classes or categories of items. For example, if Drink is a taxonomy parent of the items Juice and Milk, and the item Juice is a taxonomy parent of the item Orange juice, you can specify one of the available item constraints for the taxonomy parent Drink. This constraint affects any occurrence of the items Orange Juice, Juice, Milk, or Drink in the data.

If you want to include the specified categories, only rules that contain at least one of the specified categories or at least one of their direct or indirect members are generated. If you want to exclude the specified categories, the rules that contain the specified categories or their direct or indirect members are discarded from the result.

If you use name mappings, you can use the original item names or their name mapping to define item constraints.

You can use methods for specifying item constraints with integer ID arguments or with textual key word arguments. The following table shows the item constraints and the valid values for the integer IDs and the corresponding key words. The item constraints are sorted in alphabetical order.

Table 2. Valid values for item constraints
Item constraint	Integer	String
The item must exist in the rule body.	1	isInBody
The item must exist in the rule head.	2	isInHead
The item must exist in the rule.	3	isInRule
The item must not exist in the rule body.	-1	notInBody
The item must not exist in the rule head.	-2	notInHead
The item must not exist in the rule	-3	notInRule
At least one item of this category, or the category name, must exist in the rule body.	11	categoryIsInBody
At least one item of this category, or the category name, must exist in the rule head.	12	categoryIsInHead
At least one item of this category, or the category name, must exist in the rule.	13	categoryIsInRule
An item of this category must not exist in the rule body.	-11	categoryNotInBody
An item of this category must not exist in the rule head.	-12	categoryNotInHead
An item of this category must not exist in the rule.	-13	categoryNotInRule
Average weight of the transactions (Associations) or transaction groups (Sequences) that support the rule	29	transactionWeight

Example of item constraints

Item constraints that include an identical third argument are concatenated by the logical OR. Constraint groups that include different third arguments are concatenated by the logical AND. Therefore, the boolean predicate that matches the rule filter below looks like this:

('Roquefort' is in the rule head or 'Camembert' is included in the rule head) AND
(Any article of the class 'alcoholic drinks' or the class name itself is included in the rule head) AND
(Neither an article from the class 'alcoholic drinks' nor the class name itself is in the rule body) AND
(Neither an article from the class 'non food' nor the class name itself is anywhere in the rule).

IDMMX.DM_RuleFilter()   
   ..DM_setItemConstr('Roquefort','isInHead',1)   
   ..DM_setItemConstr('Camembert','isInHead',1)   
   ..DM_setItemConstr('alcoholic drinks','categoryIsInHead',2)   
   ..DM_setItemConstr('alcoholic drinks','categoryNotInBody',3)   
   ..DM_setItemConstr('non food','categoryNotInRule',4)

Range constraints

With range constraints, you can limit the search space for new rules by specifying valid ranges for particular numeric properties of a rule, for example, the support value or the confidence value. You can use methods for specifying range constraints with integer IDs or with textual key words.

The following table shows the valid key words and the corresponding integer IDs for specifying range constraints. The key words are sorted in alphabetical order.

Table 3. Valid values for range constraints
Range constraint	Integer	String
Accessible volume	28	accessibleVolume
Weight of the rule body	22	bodyWeight
Confidence	2	confidence
Weight of the rule head	21	headWeight
Weight of each item in rules or item sets	24	itemWeight
Lift	3	lift
Number of items	5	nbOfItems
Number of item sets	6	nbOfItemsets
Number of items in the rule body	7	nbOfBodyItems
Number of items in the rule head	8	nbOfHeadItems
Weight of the rule	20	ruleWeight
Support	1	support
Support*confidence	4	suppTimesConf
Support times weight of the rule	23	suppTimesRuleWeight
Elapsed time between adjacent parts of the rule	10	stepWindow
Total elapsed time from the beginning to the end of the rule	9	totalWindow

Example of range constraints

This example specifies that the rules that meet the following criteria are included in the model:

Support value of 2.7% or higher
Confidence value of 50% or higher
Lift value smaller than or equal to 0.5, or greater than or equal to 2.0
Consists of 2 to 4 items (counting rule body and rule head)
1 head item (more than 0 and less than 2)

IDMMX.DM_RuleFilter()   
   ..DM_addRangeConstr('support',2.7,100.0,'isInClosed')   
   ..DM_addRangeConstr('confidence',50.0,100.0,'isInClosed')   
   ..DM_addRangeConstr('lift',0.5,2.0,'notInOpen')   
   ..DM_addRangeConstr('nbOfItems',2,4,'isInClosed')   
   ..DM_addRangeConstr('nbOfHeadItems',0,2,'isInOpen')