Keras Learner isn't working for Multivariate Time Series Analysis

Dear Sir or Madame,

I have been struggling with Keras Network Learner node since days and I downloaded keras 2.3.1 version and tensorflow 2.1.0 version. I cannot downgrade keras to 2.1.6 - as Knime Console suggests me to do.

Does Keral Network Learner work only with keras 2.1.6 and tensorflow 1.8.0 versions? Below is the article I use to implement my analysis: Multivariate Time Series Analysis with LSTMs - All Codeless | KNIME

I must admit it is very uncomfortable to change between so many keras and tensorflow versions. I have installed the latest packages for keras and tensorflow and i suppose it must also work with the latest versions too.

Would you have any solution please?

Thank you in advance,
Kind Regrads,
Alina

Hi @Milovanova ,

At the moment we only test the specified versions of python packages. Some constellation of package versions may work but there is no guarantee. You should also consider that getting multiple package dependencies right is a difficult task.

Stay tuned for the upcoming release end of the year where newer python packages will be also available to be used with KNIME in general.

Best,
Temesgen

3 Likes

@Milovanova I think you would have to follow the precise Python versions and packages that are recommended for the deep learning integrations in KNIME. I found that sometimes even the repository channel you use might play a role with standard anaconda more stable than conda-forge.

https://docs.knime.com/latest/python_installation_guide/index.html#_introduction

It will be very interesting to see a better integration of newer python packages in the upcoming winter release of KNIME mentioned by @temesgen-dadi

3 Likes