Tried to get working the Deeplearning Script - only successful in small parts:

Installation: win 10, zorro 2.56, R 4.3.2, Python (version 3.11.5 or 6) within R, as alternative as system installation or Anaconda3, Keras always as R package (version 2.13), Caret, Deepnet

My Experience:
- Got working Deepnet from inside R ("neural.test()") and from Zorro as expected. Had to move set.seed() from neural.init() to neural.train() to get identical results from one to multicore training.
- Keras/Tensorflow runs within R using neural.train(). Set.seed() same as above. Had to change predict_proba() to predict(), function was removed in Keras version2.6. No chance to get it working from Zorro, used some sink() calls in neural.save(). Looks like function call Rx( neural.save....) does not start. R is still running. No idea what is wrong, does neural.train() not return properly?
- Installing MxNet impossible, URL does not exist anymore. Project has "retired"..... when you are searching for information of this project.

Due to the fact, that deepnet is limited to 3 layers (known issue, no change since years - no maintenance), MxNet not existing anymore at least for R, i would like to get working Keras/Tensorflow. In the forum i could see some comments about this problem (Keras), but no clear advise how to overcome the problem. Is it a problem of my software installation or has Keras changed so much, that the script does not work anymore (in contradiction to the working neural.test()). Is there an interference between Keras and the R-Bridge?

So any comment would be appreciated.

Hello wolfi!, I have Zorro 2.50.2, R 4.0.1 , Anaconda with python 3.8.5

I usually install these R packages: deepnet, e1071, caret and keras.

And python keras with "conda install conda-forge::keras"

I have different machines, sometimes there are missing packages or installation errors inside R and sometimes not. I normally use Keras (CPU). I have used keras GPU but couldn´t make it work much faster.

So, just to let you know that should work, but sometimes (not sure why) I had setup problems. But once it´s working, I make no changes to my machine and keep deeplearning months over months.

Hope it helps.

Best Regards

Hello tomaslolo - thank you for your comments!
Tested several installation methods, including your advise, nothing worked at the end.
Checked with debugview the keras script running. It is possible to start multiple zorro(s) for wfo in parallel, but all exit after the neural.training silently. The next step neural.save cannot open - r is not running anymore.
I do not have any idea what to do, so i will stop my activities in this issue at the moment.
Best regards!

still actual

as soon as I add `keras_model_sequential()` (and nothing else, keras installed correctly, everything runs smoothly via Rterm) to neural.train() function - Zorro fails to call neural.save()

and I think it's the same issue as mentioned in documentation: