Help with Tensorflow

[sfoglia]

Hi, I’m new to Knime and my interest is related with image recongnition, namely detecting objects in pictures.

I tried to start from the " Image Recognition for Retail Use Case" workflow but I’m blocked by the request to install Tensorflow older version 1.8.0 instead of the last updated one. Basically there’s no more any support for that version by official Tensorflow website.

Now, I’d like to ask if and when Tensorflow integration with Knime will switch to Tensorflow 2.x.

Thanks for any help and Ciao, Bruno

[bwilhelm]

Hi @sfoglia,
Welcome to the KNIME Forum!

We have multiple “Deep Learning” integrations. The workflow (this one, right?) uses the “KNIME Deep Learning - Keras Integration”. This integration is built around the standalone Keras library with a TensorFlow 1 backend. Since TensorFlow has changed significantly with TensorFlow 2 the “KNIME Deep Learning - Keras Integration” still requires TensorFlow 1. The easiest way to get a working Conda environment with the correct dependencies is the “Python Deep Learning” preference page (see our docs).

Additionally, we have a TensorFlow 2 Integration but this integration includes a new set of nodes that have to be used to make use of TensorFlow 2.

[mlauber71]

@sfoglia the point is to find the exact combination of python versions and packages that would work with the knime deep learning integrations. And they might be different depending on you use of keras, or Tensorflow 1 or 2.

The description could try and make that more obvious and also more clearly provide yaml files about what knime would want to have.

https://docs.knime.com/latest/deep_learning_installation_guide/index.html

I tried to distill what is there into Conda Environment propagations for several scenarios on windows and Python.

DL Python Network Learner

@Asa_66 you might want to try this adaption of the original with the Conda Environment Propagation with settings that should work on MacOSX or Windows: [image] @Kathrin - it might be an idea to provide YAML files or Conda Environment Propagations (like in your example here ) for the Deep Learning examples since sometimes it is hard to find the right combination that would be compatible with the KNIME version.

AttributeError: module ‘keras.optimizers’ has no attribute ‘Adam’

@akhtarameen as it happens I just have collected some example how to set up Keras and Tensor Flow 2 (for classification tasks) with the help of the Conda Environment propagation. If you have a working Miniconda or Anaconda environment configured in KNIME you should be able to just use the propagation and get a working deep learning setup. You would then have to carefully check and plan what you want to d with the setup and what architecture you want to use. [image] [image] The setup in …

mlauber71/Public – kn_example_deeplearning_keras_tensorflow_classification

@bwilhelm it would be great if the knime deep learning example could come with their own environment propagation (maybe even for Windows, Mac and Linux) and also for the deep learning guides to be more exact and maybe provide explicit yaml files for the knime nodes. Often you find circle and cross-references that can be confusing.

The latest knime version 4.6 offers a confirmation specifically for python deep learning- @sfoglia maybe you check this out.

[mlauber71]

@sfoglia Ihre one more entry about where to find the yaml files used by knime so if the (automatic) installation via the preferences does not work you could resolve to them:

Python Environment Creation Problem

@kienerj on a Windows machine you can (currently) see the YAML files used under: \knime_4.5.x\plugins\org.knime.python2.envconfigs_4.5.x.yyyyyyyyyy\envconfigs that can help to build one’s own configuration. If you already have built an environment and want to save it in a ‘clean’ (reusable) way the way to go seems to be the --from-history function: Linux … conda env export --from-history | grep -v “prefix” > environment.yaml Windows: conda env export --from-history | findstr -v “pref…