How to include all transformations as in PMML in KNIME

Question

I am at the initial stage at using KNIME, and currently I am testing the PMML export from KNIME to other languages.

I am used to export PMML from python by using the pipeline as its main component which includes all transformations, and while exporting all values of the transformations are included in the PMML file.

Problem

Now, in KNIME I tried making the classification workflow but which includes the following transformations:

• One to Many (One hot encoding)
• Column dropping (Dropping all categorical columns)
• Normalization (On the dropped columns)

And then finally the model training.

---

But when I try to export the workflow as PMML, I see I could only connect the last stage of transformation "normalization" into the PMML appender which results the file to only have the last Normalization in it.

So while running in other platform (programming language other than KNIME) it throws error that some "X" column is not found in the the available columns.

Which is because it hasn't applied the One to many step as it doesn't know that it actually exists. Here is the image of my workflow:

There, the blue region is where all transformations exist. I found a similar workflow online where the author is able to connect all transformations together which might be the solution, but in my version I can't see the option to connect transformations together (as you can see in the image) .

Here is the image of workflow from the internet:

---

My ask

As you can see the author from the internet has applied both: One to Many and Normalization but they are having 2 blue connections while I don't get that. So, will you please help me through this?

Thanking you ☺
Aayush Shah

Answer 1

I'm not sure I understand what this workflow should do but it seems that the problem is simply related to the fact that the nodes on the workflow coming from the internet are deprecated.

Deprecated Normalizer (PMML)

I think the input you are looking for has been replaced by the node "Normalizer Apply (PMML)"

Normalizer Apply (PMML)

You should be able to adapt your workflow to use this new node.