DL4J - Label column not available or not yet selected for SUPERVISED training.

[BurakIL]

Hello everyone,
I have a problem. When I want to run DL4J FeedForwardLearner (Classification) I get the problem shown in the picture.

The dataset I want to teach consists of 134 rows (classifiers) and 39 columns (variables).

I do not encounter any problem when I want to do the same operation with 40 rows and 39 columns.

But I want to teach all my dataset.
I share my model with you.
09.08.2023 DENEME DERIN OGRENME.knwf (770.0 KB)

Can you please help me?
Kind regards.

[ScottF]

Hi @BurakIL -

The way I fixed this was to add a Domain Calculator node. The clue here is the part of the message that says the “domain of the Label column might not be available” - the node is configured for the domain associated with the 40 rows, not the 134 rows. When the domain changes, the node isn’t sure what to do.

Anyway give this try, I hope it helps you.

[BurakIL]

Dear ScottF
Thank you for your correct answer. Thanks to you, I was able to teach the “learner” my entire dataset.

But I had another problem.
For a classifier, the predictions are correct when I give three or less values, while the predictions are wrong when I give more than three values.

Do you have any suggestions regarding this?

Thank you for your help.

[ScottF]

I’m not sure I understand what you mean. Can you upload the latest version of your workflow? How do you determine which predictions are incorrect, and how are you scoring the results of your model?