Failed to run Deep Learning examples

[anguslou]

I am trying to run the DL examples but both failed

1. 08_Sentiment_Analysis_with_Deep_Learning_KNIME…

Error as following:
ERROR Keras Network Learner 0:16 Execute failed: An error occurred while creating the Keras network from its layer specifications. Details:
Deep learning Python back end installation tests failed.
Cause: Python library ‘keras’ or one of its dependencies is missing.
Further output: Using TensorFlow backend.

In case Python packages are missing: you can create a new Python Conda environment that contains all packages required by the KNIME deep learning integrations in the “KNIME Deep Learning” Preferences.

2. 07_GAN_MNIST_Digit_Generator…

Error as following:
ERROR DL Python Network Learner 3:98:0:24 Execute failed: An error occurred while creating the Keras network from its layer specifications. Details:

Deep learning Python back end installation tests failed.

Cause: Python library ‘keras’ or one of its dependencies is missing.

Further output: Using TensorFlow backend.

In case Python packages are missing: you can create a new Python Conda environment that contains all packages required by the KNIME deep learning integrations in the “KNIME Deep Learning” Preferences.

My machine: Mac mini M1
My Python Deep Learning setting by “New environment” so it is anaconda to create:

And both Keras and TF2 extensions are installed:

Please help.

[mlauber71]

@anguslou I ave prepared these examples with Conda Environment Propagation

Train simple CNN – mlauber71

Train a simple Multilayer Perceptron using TensorFlow 2 – mlauber71

You might want to try these Conda Environment propagations (here TensorFlow2)

a simple deep learning network for binary classification using the KNIME...

You will finde YAML configuration files integrated in the sample workflows:

You will have to follow the instructions to create a compatible Python environment for Deep Learning that is suitable for your operating system.

You might want to read about KNIME and Python to understand how this works. Unfortunately the standard installation does not always work.

Keras and TensorFlow 1 and TensorFlow 2 would need different combinations of packages in order to work with KNIME.

Meta Collection about KNIME and Python – mlauber71

[bwilhelm]

Hi @anguslou,
The example workflow uses Keras and TensorFlow 1. However, there seem to be problems with TensorFlow on the M1 Mac. There are special versions of TensorFlow 2.5 and 2.6 for the M1 Mac (Tensorflow Plugin - Metal - Apple Developer) but not for TensorFlow 1.
According to this GitHub issue comment the problem seem to be that Rosetta 2 does not support the “AVX” instruction set which is used in the precompiled TensorFlow binaries. The user managed to get it running by compiling TensorFlow himself (which is not an easy task).

[anguslou]

I tried both the MacOS Conda Environment Propagations and don’t work. Error message for Python 3.6 as following:

ERROR DL Python Network Learner 4:98:0:24 Execute failed: An error occurred while creating the Keras network from its layer specifications. Details:

Deep learning Python back end installation tests failed.

Cause: Python library ‘keras’ or one of its dependencies is missing.

Further output: Using TensorFlow backend.

In case Python packages are missing: you can create a new Python Conda environment that contains all packages required by the KNIME deep learning integrations in the “KNIME Deep Learning” Preferences.

[anguslou]

Hi bwilhelm,

Thank you for your reply. Do you see it will be fixed in the coming future? Or should I move to windows instead?

[kienerj]

It will be fixed some day as apple is moving away from x86 and they will have to fix it but it could be months or years. Right now for data science (python) stuff you need for work I would not bother with these machines yet. Note that macs work entirely fine as long as they are x86 (intel) based macs.

[anguslou]

Got it, thanks…so can i say everything in KNIME is supported by M1 except deep learning?

[mlauber71]

@anguslou have you tried to set up tensorflow2 with the special packages suggested by @bwilhelm? Would be interesting to see if they would work.