I have noticed a strange bug with Database groupby

KNIME Analytics Platform
#1

Hi collegues, 

Here above you can see an example in which I convert data in a transactional format and then I process it into Borgelt node for extract rules (from node 11 to 28). If I convert to table the database and then I convert them in a transactional format with groupby node, then the Borgelt extract correctly the rules.  

The strange thing happens when I try to perform the same operation to convert data with Database groupby node and Create Collection Columns node(from node 18 to node 27). I got the exact result as data format (an aggregated list of values) that I have also with the simple groupby. I have also compared the data transaction obtained together and they match as rows values and numbers. The data are exactly the same.

In the first working case I got as result about 23.000 rules.

In the second case It just creates an empty table with no other explanation in knime Console.

Can someone clarify me why the first procedure work and the other not?

Thanks in advice.

-Giulio

#2

Hello Giulio,

what database are you using? KNIME does not support reading of collection columns from all databases since the array implementation is different for each database.

To check that have a look at the specification of the KNIME table the Database Connection Table Reader creates. To do so right click on the Database Connection Table Reader node and click on the last entry "Data read from Database". In the dialog go to the Spec tab and check that the column you aggregated is of type Collection.

Bye

Tobias

#3

Hi Tobias,

you are right, unfortunately I didn't realized that group concat for MySQL was not the same thing of Collection type that is the input data type required by the Association Rule Learner.

Since the format at a first look seemed the same (like an aggregated value) I thought it could be used anyway.

-Giulio