We need a new kind of join tool - Any Takers?

Hi Gavin,

well your example shows the solution. NULL /missing value is simply something entirely else than a blank string. I think you need to read up on what null values are. For what you want, the missing value node is the solution and that should under no circumstance be added to the joiner. If for you null is equal to blank string, then use the missing value node. Problem solved. I really fail to see why this should even be in the joiner. It does absolutely not belong there
.
KNIME isn’t about providing these huge black-box like feature blocks. It provides you basic building blocks (with simple configuration) from which you can build your own “algorithms”. For me the only relevant change to Joiner would be performance increase. The rest you want can easily be done with components and hence don’t belong into the node itself.

3 Likes

@beginner
So, having a bit of free time this evening i decided to venture into some component building to familiarise myself with the process and thought on your prompt why not try and build the join tool i am after.

getting the outputs and that config was easy enough, however the pickle seems to come when designing the components interface, in particular in selecting the join columns. See i’m not working to the assumption that the cols have the same name let alone the same order. I can’t seem to find a configuration option that allows me to select columns in a particular order.

now i know there are workarounds of re-naming prior to input of the component and sorting the column order but that would mean anything from 1 to 4 nodes in total in addition to the component, might as well not use a component in the first place…

any thoughts?
oh and how would we re-build the column selection interface for the inner output?

1 Like

Hi @Gavin_Attard,

so basically which columns are used for the table join?
Maybe you could just add the column selection node in 2-Pairs for each join key? :thinking:

So for one join condition 2 of these nodes
and for two join condition 4 :thinking:
Maybe someone has a better idea :slight_smile:

Thanks @AnotherFraudUser - ye i considered that approach, limiting to say 3 join columns.
I think then the coloumn selection to pass through can be (to a certain extend) mimicked also.

1 Like

Hello there!

guess have seen that the new Joiner node - Joiner (Labs) is available from KNIME version 4.2.0. It is still in Labs but outputs unmatched rows, much faster and handling missing values pretty well. In case you haven’t here is more info: https://www.knime.com/whats-new-in-knime-42#joiner-node

Br,
Ivan

4 Likes

This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.