I have just executed an example use case from the KNIME book - Codeless Deep Learing with KNIME
The case is Sentiment Analysis Training with KNIME and the accuracy result I get is 10% lower than that shown in the book. Does anyone know why that might be?
Reliability of results (replicablility) is important or at least interpretability.
Great to hear that you are practicing with the book!
I’m sorry for the mistake in the referenced workflow. I ran the workflow (Sentiment Analysis with LSTM - Training – KNIME Hub) and only achieved an overall accuracy of 69%, so the same as you.
I assume that we made a mistake when adapting the workflow for the book chapter, because the original workflow (Sentiment Analysis – KNIME Hub) gives the same results as those reported in the book.
Thanks for reporting the issue. I hope we find the reason soon, and we will correct the other workflow accordingly.
Thank you for looking into this. As a new user the first reaction is to assume that I have made a mistake.
Will you be able to send me the corrected workflow please?