Announcing KNIME Analytics Platform 5.2

[takbb]

Hi @armingrudd , I’ve also been seeing the erratic node behavior in 5.2. To me it’s a little different to 5.1 but I’ve yet to be able to work out how to make it happen.

It happened to me a couple of times over the weekend. When it next happens I’ll try to record it so you can see it in action even if i can’t find the steps to reproduce.

[iCFO]

I am also a very heavy user of the keyboard shortcut for opening the output view port table. I run a dual monitor rig (or DisplayFusion to split screens when on a laptop). The keyboard shortcut for the output table allows me to use a macro to open output tables, move it to the next monitor for copy and paste access into config windows, then open the config window of the selected node where I will write expressions.

I use it constantly to copy / paste values and get around being locked into config windows. Zoom in (Ctrl +) still isn’t working on my end either.

I also really miss the keyboard shortcut area in preferences where shortcuts can be changed and viewed easily. It really simplified finding all of the currently available keyboard shortcuts and avoiding overlap between programs. I know that keyboard shortcuts are still a WIP, but they would allow us to sidestep issues and maybe take some pressure off of UI development.

[iCFO]

I am able to work in the new UI much more frequently thanks to this v5.2 update. Great work KNIME team!

[TosinLitics]

Hello hello guys!

I love the AI nodes and I can see this being a massive game changer for many businesses.

I am working on a RAG implementation using Python, but I would love to switch the solution over to KNIME. I have outlined below a few missing functionalities on the KNIME side that are roadblocks for me to taking a no-code approach:

1. The ability to ingest metadata to the vector store and to filter by these metadata when retrieving.

update - looks like the ability to add metadata has already been added to the nodes already

2. In addition, when creating a vector store it would be ideal to create a collection where more documents can be added on demand in the future.

3. Vector database management nodes that allow for the deletion of certain collections, or documents within that collection.

Here are some wishlist items:

1. It would be nice to have a node that handles the chunking of large texts

2. Another nice to have would be a cross-encoder re-ranking node to ensure that the most relevant results are at the top.

3. A way to visualize where the returned docs lie in the vector space, relative to the query.

I’d be happy to discuss this further, and please let me know if some of these functionalities are already available…maybe I am just missing something.

[mlauber71]

@TosinLitics I do not have extensive experience with Document metadata but you might try to add them using the Document Data Assigner and provide them to the FAISS Vector Store Creator. Maybe it is worth experimenting with that.

(from this example)

Creating a Local LLM Vector Store from PDFs with KNIME and GPT4All

https://hub.knime.com/-/spaces/-/~WNe6bb2w2bemYBWE/current-state/

Also you could experiment with the ‘window’ size of texts and documents you are providing:

correct relative path for GPT4All workflow (GPT4All - Create your own LLM Vector Store)

@mgirdwood I have this example where I load several PDFs from the author Stefan Zweig into a Vector store and make them accessible. You might have to experiment with the meta data of the documents and also with the ‘window’ you give the model. There is a component that tries to do a more complex preparation of the data with giving blocks of text (since a novel might benefit from a few sentences of context). You might have to toy around with such concepts for your own data. A coffee machine ma…

[TosinLitics]

This looks promising. I will try it, thank you!

[wpaiva]

It would be nice to have a shortcut to open the table, instead of going through the path Open output port > Table.

[takbb]

Hi @wpaiva , you may wish to add your vote here if you haven’t done so

View data by shortcut

Hi I would appreciate, if the data of out ports could be viewed by having selected the node and hitting a number key according to the ordinal from above, e.g. for the first out port hit 1, for the second hit 2, … . I presume, that out port number > 9 will be rather seldom. Kind regards Thiemo