如何在R中使用keras使用Python安装的tensorflow

Question

I have already performed the tensor flow installation with the following command:

pip install --ignore-installed https://github.com/mind/wheels/releases/download/tf1.5-gpu-cuda91-nomkl/tensorflow-1.5.0-cp27-cp27mu-linux_x86_64.whl

This is the latest tensorflow wheel catered for CUDA 9.1. (3x faster than CUDA 8.0)

And I can call it successfully in my python code.

How can I make the keras in R to call the tensorflow installed by python above? The reason I asked that because I the default installation method

keras::install_keras(method="conda", tensorflow = "gpu")

It failed to recognize the cuda-9.1 library.

> conv_base <- keras::application_vgg16(
+   weights = "imagenet",
+   include_top = FALSE,
+   input_shape = c(150, 150, 3)
+ )
/home/ubuntu/anaconda2/envs/r-tensorflow/lib/python2.7/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters
Using TensorFlow backend.
Error: ImportError: Traceback (most recent call last):
  File "/home/ubuntu/anaconda2/envs/r-tensorflow/lib/python2.7/site-packages/tensorflow/python/pywrap_tensorflow.py", line 58, in <module>
    from tensorflow.python.pywrap_tensorflow_internal import *
  File "/home/ubuntu/anaconda2/envs/r-tensorflow/lib/python2.7/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 28, in <module>
    _pywrap_tensorflow_internal = swig_import_helper()
  File "/home/ubuntu/anaconda2/envs/r-tensorflow/lib/python2.7/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 24, in swig_import_helper
    _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
ImportError: libcublas.so.9.0: cannot open shared object file: No such file or directory

This is because R installation method calling for tensorflow version 1.5.0 that is still not catered for CUDA 9.1.

Answer 1

Try to put this in your .bashrc :

export KERAS_BACKEND='tensorflow'

Or based on this instruction , you can do:

tensorflow::install_tensorflow(version = "https://github.com/mind/wheels/releases/download/tf1.5-gpu-cuda91-nomkl/tensorflow-1.5.0-cp27-cp27mu-linux_x86_64.whl")

Then keras will automatically identify the correct tensorflow

Answer 2

R is looking for the CUDA 9.0 version, rather than the latest 9.1. You should be able to symlink your system so it ends up at the 9.1 folder instead; something like:

ln -s [path to cuda 9.0 it's looking for] [cuda 9.1]

Alternatively you may be able to uninstall 9.1 and install 9.0. I believe you will also need cudnn version 7.