TensorFlow + KNIME Deep Learning Plugin: Environment Setup Fails Due to Archive and Compatibility Issues

Hi KNIME team and community,

I’m running into persistent issues while setting up a Conda environment for the KNIME Deep Learning Keras Integration (KNIME 5.4.0 on Windows 10). My goal is to use TensorFlow 2.10 with Python 3.9, which I understand is required for compatibility with the KNIME deep learning extensions.

Here’s a summary of the issues I’m encountering:


:exclamation: Key Issues:

  1. Broken TensorFlow Archive:
    When creating the environment with:

    conda create -n knime-tf python=3.9 tensorflow=2.10
    
    
    I get this error:
    
    
    InvalidArchiveError: Error with archive tensorflow-base-2.10.0-mkl_py39...
    You probably need to delete and re-download or re-create this file.
    
    
    Even after manually deleting the `.conda` file, the error persists during reinstallation.
    
    
  2. Unhashable Tensor Error in KNIME:
    After attempting to run the Keras network inside KNIME, this error appears:

    TypeError: Tensors are unhashable …
    Instead, use tensor.ref() as the key.

This occurs during execution in the DLKerasNetworkSpecExtractor.py module used by KNIME internally.

Outdated pandas.util.testing reference
An earlier run failed due to this internal import in DLPythonNetwork.py (plugin file from KNIME):

ModuleNotFoundError: No module named ‘pandas.util.testing’

This module has been removed in recent pandas versions, so the KNIME plugin might need updating or patching.

  1. Python Version Conflict
    An attempted environment with Python 3.13 (by mistake or default) caused conflicts with TensorFlow 2.10, which only supports up to Python 3.9.

My Setup:

OS Windows 10
KNIME Version 5.4.0
Miniconda Latest
Goal: Run TensorFlow 2.10 + Python 3.9 in KNIME for deep learning workflows

Could you please provide a working and officially supported configuration, including:

  • Compatible versions of Python and TensorFlow
  • Required Conda packages and versions
  • Potential patches for deprecated pandas code in DLPythonNetwork.py
  • Ideally a working environment.yml file or reproducible setup instructions

Hi @R2D2 and welcome to the forum. (I’m a little surprised someone hasn’t taken that name already!)

Take a look at the last couple of posts in this thread, particualrly @mlauber71 's, and see if that helps at all:

2 Likes