Error with Keras Deep Learning Network - Learner fails execute, issue possibly related to Python DL Configuration

[JoshuaMMorriss]

I am running into an error where my Keras Network Learner will not execute. It will get stuck on “Executing - Copying input object at port 1 - Shuffling training data - Merging temporary tables,” “Executing - Processing batch 1 of 127 and epoch 1 of 69…”, then error out with the following message below.

I also get the following warning message:

WARN Keras Network Learner 0:16410:16360:148 C:\Users\my_username\Anaconda3\envs\py3_knime_dl_1\lib\site-packages\keras\engine\saving.py:292: UserWarning: No training configuration found in save file: the model was not compiled. Compile it manually.

Below was the configuration of my Keras Network Learner node, the tuple was a flow variable with a default of 10 steps and x number of dimensions. The dimensions are divisible by 10, and are based on the number of features used in the entire matrix (i.e. 130 columns/features, so the structure would be [10,130]). I am also tuning the flow variables (hyperparameters) using a Parameter Optimization Loop, although none of these variables seemed to cause the error in my prior executions of the Keras Learner Network.

I have tried different configurations of my python deep learning environment, using Conda as my environment. One of the earlier configurations I used was this:

Below is the current configuration I have:

My current version of KNIME is version 4.5.2, and the Keras-relevant KNIME extensions are as follows:

1. KNIME Deep Learning - Keras Integration (Version 4.5.0)
2. KNIME Deep Learning - TensorFlow 2 Integration (Version 4.5.0)
3. KNIME Deep Learning - TensorFlow Integration (Version 4.5.0)

I am not sure how to traceback this error. If it is a problem with the tuple (and thus, with the train/test data matrices), a problem with the python deep learning environment, a problem with any of my KNIME extensions, or an entirely different cause. Any feedback at all would be greatly appreciated, thank you!

[JoshuaMMorriss]

Hello! I was able to determine that this error was associated with improperly configured Keras / TensorFlow 2 Conda environments. Originally, I was using GPU environments instead of running everything on my CPU.

With the newly created environments below, I no longer encounter this error.

To reconfigure from a GPU environment to a CPU environment, simply select “New environment…” and ensure that you select Create new CPU environment.

[ScottF]

Thanks for following up and posting your findings!