python deep learning error — tensorflow version<=2.0（Mac OS，M4）

[Guo_Yafei]

new environment error

Subject: Configuration Issue: “TensorFlow 2” Option Requires Version <2.0? (macOS M4)

Hi,

I’m having a persistent issue setting up the Python Deep Learning (Keras with TensorFlow 2) integration on my Apple Silicon Mac (M4 chip) and believe there might be a configuration bug or validation logic issue.

1. My Environment & Setup:

• Computer: Mac with Apple Silicon M4.

• KNIME Version: 5.8

• Python Management: Miniconda (conda 25.1.1).

2. What I Did (The Correct Path):

1. In KNIME, I navigated to: Preferences → KNIME → Python Deep Learning.

2. I selected Use special Deep Learning configuration.

3. Under Library used for the "DL Python", I chose Keras and made sure the sub-option TensorFlow 2 was checked.

4. I set the configuration method to Conda and pointed it to an existing environment.

3. The Core Problem (Contradictory Error):
The environment validation fails with a confusing error that directly contradicts my selection:

• It correctly detects that I have TensorFlow 2.16.2 installed.

• However, it reports: required minimum version is 1.0.0 (inclusive), required maximum version is 2.0.0 (exclusive).

This is the key issue: I explicitly selected “TensorFlow 2”, but the system is checking against the version range for the legacy “Keras (TensorFlow 1)” integration.

4. What I Have Tried (And Failed):

• Manual Lower-Version Environment: Created a Conda env with Python 3.8 to try installing older TensorFlow. This failed because pre-built packages for TF <2.0 are not available for osx-arm64.

• KNIME’s Auto-Create: Clicking the New environment... button in the config panel also fails due to unresolvable dependencies for the Apple Silicon architecture.

5. My Question / Suspected Bug:
Is this a known issue where the wrong version validation logic is triggered when the “TensorFlow 2” option is selected on macOS? The error message strongly suggests the backend is applying the checks for the old Keras integration, making it impossible to pass.

Any insights or workarounds would be greatly appreciated. Thank you!

[mlauber71]

@Guo_Yafei for Apple silicon there seems to be a workaround:

Using tenserflow version 2

Hello @Kyzzhibek, So apparently there is a workaround, but it is not well documented yet and it is in the process of being formally documented for the workaround. From looking at the documentation I believe you will need to do the steps below to create the conda environment: Download the text file: knime_dl_mac.txt (10.3 KB) In terminal run the command below to create the env: conda create –name knime_dl_mac --file knime_dl_mac.txt Hope this helps, TL

Other than that has deep learning and knime always been a challenge. I wrote this article how to approach the problem.

https://medium.com/low-code-for-advanced-data-science/knime-and-python-setting-up-deep-learning-environments-for-keras-and-tensorflow-4b66003858f4