Hello, 

thank you for the feedback. Unfortunately I do not have any ready log to post.

I am trying to extend Mahout library with another MapReduce step to compute the 
association rules.
I can form the association rules from Mahout's output, but I cannot compute the 
interestingness measures such as confidence.

For instance, let's say that we have a frequent pattern as output [a,b,c,d,e]. 
We already know the pattern's support. The idea is to form a rule by removing 
one item and put it at the right part of a rule, for example: abcd->e .

The confidence measure is computed by dividing the support of the [abcde] item 
set that we already know by the support of the item set [abcd] which is the 
right part of the rule. But we do not have this information because the item 
set [abcd] may not even be a frequent one and it is not feasible to search all 
the patterns for the computation of one association rule.

Any suggestions are more than welcome.

Thank you in advance,
Konstantinos
On May 13, 2013, at 12:48 PM, Louis Hénault <[email protected]> wrote:

> Hi,
> 
> After using FPG within Mahout, you get raw results like this:
> 
> Key: i: Value: ([i],K1), ([i, j],K2), ([l, i],K3), etc...
> 
> This means that for the key i, the top associations with i are first i
> in K1 transactions, then i and j appears in K2 transactions, etc...
> 
> with K1 > K2 > K3 > ...
> 
> Then, you have to use the raw results to measure your associations
> rules. You can use several metrics, e.g conviction measure (see
> http://en.wikipedia.org/wiki/Association_rule_learning#Alternative_measures_of_interestingness
> ).
> 
> I hope it helped,
> 
> Louis
> 
> 
> 
> 2013/5/13 Konstantinos Patsakis <[email protected]>
> 
>> Hello,
>> 
>> I am using Mahout's FP-Growth algorithm an I would like to extract
>> association rules after the frequent pattern mining.
>> 
>> However, I am experiencing some difficulties because I do not know how to
>> manipulate the output of Mahout to generate association rules. Maybe you
>> could help me with this or you need more info about my obstacles?
>> 
>> Thank you in advance,
>> Konstantinos

Reply via email to