Hello everyone.
Regarding the R package arules. I've realized that (at least) two of the most common measurements for APRIORI are actually using intersection instead of union.
Support, supp(A ∪ B ) the proportion of the transactions that contain both itemsets, is actually implemented as A ∩ B.
Confidence, supp(A U B) / Supp(A) is implemented as Supp(A ∩ B) / supp(A). I'm not sure, but seems that Weka is implemented using the intersection as well.
Ex:
# The following input:
A
Hi Joaquim,
The Apriori algorithm in WEKA would produce the following detailed rules from your data:
1. Z=0 4 ==> A=0 4 lift:(1.43) lev:(0.12) [1] conv:(1.2)
2. A=1 3 ==> Z=1 3 lift:(1.67) lev:(0.12) [1] conv:(1.2)
Each rule in the output shows two counts: the support for the full rule is shown on the right-hand side of the rule; the support for the left-hand side is shown just before the implication symbol. Confidence is the former divided by the latter. In Apriori, confidence is simply the estimated accuracy of the rule (i.e., percentage of the training instances the rule associates correctly).
However, another option one can consider is applying the FP-growth algorithm to deal with your data which produces this rule:
[A_binarized=1]: 3 ==> [Z_binarized=1]: 3 lift:(1.67) lev:(0.12) conv:(1.2)
HTH.
Samer
Thank you Philippe. The union is kind of trick, I'm not sure that most of people really understand it... I'll check SPMF and read you survey as well.
Thank you Samer for your answer.
- Badges
- Science topic