I am new in Knime and I am probably going to ask something very naïve, so my apologies in advance.
I am trying to implement a recommender application. I am prototyping it by using the Association Rule Learner (Borgelt) and I am confused about how to handle the itemset support variable, I assume it has to be defined in % through the minSupport variable in the Double Configuration node linked to the ARL(B) node.
But then, inside the ARL(B) node, there is another variable to configure in “Item set settings” also called Minimum Support and allowing to select Absolute Number or Percentage, which I don’t really manage to figure out what it is for, as it is not completely ignored (e.g. it alters the output in the case you input an Absolute Number).
Additionally, when I run the learner, I am expecting to get rules from itemsets with a support over minSupport, but the ItemSetSupport column is not showing so, it is the RelativeBodySetSupport% the column fulfilling that rule (>minSupport, 2.5 in this case).
@Luisa_introuble welcome to the KNIME forum. The confusion might arise from the use of variables (that is “Flow Variables” in the case of KNIME).
The “Double Configuration” would create a Flow variable called “minSupport” and this variable is then fed into the Rule Learner; the use of the variable is indicated at the bottom “… is controlled by a variable” - this means the Minimum Support of 10 which you see is in reality the 2.5 you have configured. You can see this on the “Flow Variables” tab. The variables would overrule what is present in the node configuration.
Thank you very much @mlauber71. May you confirm whether this minSupport is in % over the total number of transactions used to train the learner? If so, the number of transactions in the screenshots I posted above is 2869, being 72 transactions the 2.5%. But the ItemSetSupport in the Association Rules table is starting in 8, and the one starting in 72 is the AbsoluteBodySetSupport. This it seems that the minSupport threshold is being applied only over the frequency of the itemsets in the antecedent of the rule, not over the full itemset (as I was expecting from the apriori algorithm). I do appreciate any explanation to this. Thanks again!