Using keras with tensorflow-rocm

[chevrox]

I'm trying to use an AMD GPU for deep learning and I was able to successfully install tensorflow-rocm using either the tensorflow or keras package installer:

install_keras(tensorflow = "...path/to/tensorflow-rocm-package.whl")

or

install_tensorflow(version = "...path/to/tensorflow-rocm-package.whl")

obtained from https://pypi.org/project/tensorflow-rocm/#files

However, but when I tried to use keras after installation I get the following error:

Error: Installation of TensorFlow not found.

Python environments searched for 'tensorflow' package:
 /home/chevrox/.local/share/r-miniconda/envs/r-reticulate/bin/python3.6

You can install TensorFlow using the install_tensorflow() function.

Help appreciated.

[phil_hummel]

Welcome to the R Studio Community!

@chevrox

I'm trying to use an AMD GPU for deep learning

All the keras for R docs I find specify you must have an NVIDIA GPU for this package. Do you have any doc references that state AMD GPUs are supported?

[chevrox]

Nope, but I had hoped that as long as a rocm backend is specified and installed it would work as usual. After all, keras does not use a customized version of tensorflow, does it? Since tensorflow-rocm has the same syntax as tensorflow, I was hoping that it would work.

[phil_hummel]

R interface to Keras

R Interface to 'Keras'

This website provides documentation for the R interface to Keras. See the main Keras website at https://keras.io for additional information on the project.

There are differences between the R interface and the Python API for Keras.