Deep Learning Keras Installation

[AnthonyCREng]

Greetings Knime Community.

I woulkd like to know if anyone is able to assist with the following case:

I need to install Tensorflow and Keras packages for Depp Learning affairs.

Initially I have installed the following extensions as per TSA algorithms demands:

1. KNIME Python Integration
2. KNIME Deep Learning - Keras Integration.
3. KNIME Deep Learning - TensorFlow Integration.

Additional information:
Knime Analytics Platform Version: 4.6.4
Python version: 3.9.13
TensorFlow Package Version: 2.9.0
Keras Package Version: 2.9.0

Keras Version does not fit due to TensorFlow version package (2.9.1) which exceeds the one required for a proper functionality (Knime suggests versions from 1.0.0 until 2.0.0).

Based on the prior explanation I decided to downgrade Tensorflow package applying <pip upgrade tensorflow==2.0.0> however python didnt applied this action indicating impossibility to rewrite tensorlow, hence I had to uninstall Tensorflow; subsequently I reinstall Tensorflow again but this time using version 1.15.0 notwithstanding Anaconda Powershell shows and indicates impossibility to accomplish this action, please se the error below:

I am trying to install the sugggested tensorflow package version and so utilize keras, however KNIME Analytics Platform doesnt accept the latest tensorflow version and cant install an old tensorflow package version (from 1.0.0 until 2.0.0), so I am not really sure what exactly can I do.

Does anybody can provide technical criteria about how to solve this scenario?
Thanks in advance.

Quick View:

[mlauber71]

@AnthonyCREng you could take a look at this

Error please need help - Error: Network learner Error

@Haroon2001 I have a longer article coming up about KNIME and Deep Learning integration. In the meantime you could check out this entry: Or try to use these Keras and TensorFlow examples with Conda Environment propagation after you have set that up (KNIME and Python — Setting up and managing Conda environments | by Markus Lauber | Nov, 2022 | Medium).

Tensorflow/Keras installation conflict

@TargetFilled what you can do besides following the KNIME Deep Learning Integration Installation Guide and letting KNIME do the deep learning setup (KNIME Deep Learning Integration Installation Guide) is getting the official KNIME yaml configuration files for deep learning and use them just with miniforge and conda-forge repository to avoid licencing problems and also have a clean and slim setup only settings the minimum restrictions with versions KNIME is recommending: Or you can try some de…

[AnthonyCREng]

Alright got it, I am going to revise the information provided, let me try and evaluate results, then I will come back to you to let you know the outcome.

[mlauber71]

@AnthonyCREng if you want you can now check out my longer article about property setting up KNIME and Python and Deep Learning environments:

KNIME and Python — Setting up Deep Learning Environments for Keras and TensorFlow

The follow-up to my article about KNIME, Python and Conda is now online dealing with the question of: KNIME and Python — Setting up Deep Learning Environments for Keras and TensorFlow

[AnthonyCREng]

Thanks for sharing me this fascinating summary @mlauber71 Let me tell you that I have resolved the issue regarding Deep Learning Installation extensions (Python, Keras, Tensorflow, Conda-Forge)

Steps:

1. Downloand Anaconda
2. Create a DeepLearningEnvironment via Anaconda Prompt.
3. Install the following extensions in Knime Analytics Platform:
   03.01 Keras
   03.02 Tensorflow
   03.03 Python Extension
4. Set Preferences/Knime/Python/Python(Labs) = BUNDLE
5. Set Preferences/Knime/Python/PythonDeepLearning, as follows:
   
   

Voila! Ecco! Listo!

[Fernivel]

Thanks a lot gentlemen. Some stuff is to tough for me, but I will try to understand.

I share this topic from stackoverflow that also helped me in finding the solution (I am an IT dummy)

Install TensorFlow with specific version on Anaconda

[AnthonyCREng]

Thank you, @Fernivel, for your words and feedback. You are right, this is an alternative method we could pursue with conda-forge. Thank you for your contribution to this discussion.