20220716 Pikairos How to convert normalized to predictions

Please find here a possible solution based on the “RProp MLP Learner” node.

I have a few comments concerning this implementation based on the “RProp MLP Learner” node:

The “RProp MLP Learner” node is quite limited and I would rather use the KNIME Keras nodes to implement Neural Networks. For instance, one cannot chose what input variables to inject to the MLP NN and the MLP NN output has to be normalized between 0 and 1. This is not the best choice to predict an unbounded double output variable.
This somehow obliges to use the 0-1 min-max normalization rather than the Z-score (Gaussian) Normalization. This latter is a much better choice … just to mention a few drawbacks.

Nevertheless, I have completed this workflow and I’m providing here a possible solution.

Last comment, please be aware that the “Normalization Model” should be calculated only from the Training Set to avoid any data leakage from the training set to the test set. Using the whole data set to calculate a normalization before data partition leads to data leakage. Hence, this is what the uploaded workflow is doing.

Hope al this is clear enough and of help. Otherwise please reach out for further explanations.

External resources

• How to return normalization to predictions
• How to return normalization to predictions