Platform: 3.5.3
I have a workflow that has a backwards feature elimination routine which leverages a gradient boosted tree regression model. This feature selection loop is in a metanode and used in a higher level workflow, inside of a loop. The workflow executes as expected, nothing wrong there. However, when I restart KNIME and attempt to load the workflow, I am presented with an error that points to the predictor node of the feature selection loop, stating that the string for key "predictionColumnName" was not found
. I am not using any flow variables for the feature selection nodes. This error occurs regardless of whether the prediction column name is set or left alone via the “Change prediction column name” checkbox.
A possible side-effect of this error might be causing KNIME to incorrectly load some parts of the UI, which causes a lot of confusion for me when I try to reconfigure nodes.
For instance:
The feature selection end node is normally set to minimize the selection metric; however - after restarting KNIME, and possibly due to the error in loading the predictor within the feature selection loop - the minimize check box is not checked “on”, as intended and saved in the workflow. The workflow executes as expected. Therefore, the issue is likely the UI not loading the settings properly.
I am only speculating that these two issues might be related. They may be entirely separate issues.