Bug Report: Node "Association Rule Learner (Borgelt)"

KNIME Analytics Platform
1

Dear KNIME-Team,

when I apply the node to a dataset with many transactions and a few items some values in the "RuleLift%"-column are negative while the values in the "RuleLift"-column are plausible (see attached jpg).

Best regards
Martin

PS:
Knime Version 3.1.1
Windows 7, no further details accessible due to shrinked user priviliges

 

2

Hello Martin,

thanks for reporting that issue. We tryed to reproduce it but failed. Can you give us some more information about the data or a subset of the data to reproduce the problem.

Thanks
Tobias

3

Hello Tobias,

sorry for the delayed reply, I am quite busy at the moment.

Unfortunately I could not provide access to the original data since these belong to my company. Nevertheless, in the meantime I figured out some circumstances when this behaviour occurs:

1) A large fraction of the transactions is identical.

2) The total number of trancsactions exceeds 300.000. Things becomes worse if the number of transactions is increased.

Maybe you can simulate a transaction database using these information to reproduce the error. If I figure out further details I will let you know.

Best regards
Martin

4

Hello Martin,

sorry for the long delay. Thanks to the information you provided we could reproduce the problem. It seems the problem is fixed in a newer version of the association rule algorithm which we will include in KNIME after evaluation.

Thanks for

Tobias

5

Hi Martin,

we have updated the apriori binary (version 6.22) and this fixes the problem. The fix will be distributed as part of KNIME version 3.2.2.

Thanks, 
  Bernd
 

6

What is the likely deliver time for 3.2.2 Are we talking weeks, months, some significant portion of a year?

Thanks for any insight you can share.  I understand it you can not.

7

Non-official response: Weeks to a few months at most, as 3.3.0 will be released on 06-Dec-2016 (the one fixed annual release date, because tradition). :-)

-E