Keras: Error: Python module was not found

Created on 16 Aug 2017  Â·  34Comments  Â·  Source: rstudio/keras

I got keras with tensorflow to install on R 3.4.1 via:

devtools::install_github("rstudio/keras")
library(keras)
system("conda config --set ssl_verify false")  # get past conda sslerror's
install_keras(tensorflow = "gpu")

However now when I try to run the R sample code it's throwing Python modules not found when attempting to load the mnist dataset:

> library(keras)
> mnist <- dataset_mnist()
Error in py_module_import(module, convert = convert) : 
  ModuleNotFoundError: No module named 'tensorflow'
In addition: Warning message:
In stop(e, call. = FALSE) : additional arguments ignored in stop()

So I loaded tensorflow and tried telling it the correct conda environment manually but still gets errors for keras:

> library(tensorflow)
> use_condaenv('r-tensorflow')
> mnist <- dataset_mnist()
Error: Python module keras was not found.

Detected Python configuration:

python:         C:\Users\xyz\Desktop\work\software\Python\Conda3\python.exe
libpython:      C:/Users/xyz/Desktop/work/software/Python/Conda3/python36.dll
pythonhome:     C:\Users\xyz\Desktop\work\software\Python\Conda3
version:        3.6.0 |Anaconda 4.3.1 (64-bit)| (default, Dec 23 2016, 11:57:41) [MSC v.1900 64 bit (AMD64)]
Architecture:   64bit
numpy:          C:\Users\xyz\Desktop\work\software\Python\Conda3\lib\site-packages\numpy
numpy_version:  1.11.3
keras:          C:\Users\xyz\Desktop\work\software\Python\Conda3\lib\site-packages\keras

python versions found: 
 C:\Users\xyz\Desktop\work\software\Python\Conda3\python.exe
 C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\env_full\python.exe
 C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\r-tensorflow\python.exe
 C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\tensorflow\python.exe

Note: I also have another conda environment set up running keras/tensorflow fine via python code, not sure if there's some issues with the use_condaenv() that it's not detecting correctly.

Thanks for any advice.

picture wtcooper
👍1

Most helpful comment

This issue seems to be reappearing from time to time every month on the same machine & same Rstudio. I have started dreading this error "Python module not found". Every time I forget what was the solution and I start to search and land on this GitHub thread. Waste a lot of time reading the entire thread eliminating each possibility - painfully & well explained.
But at the end, just a restart of Rstudio works.

I think there is some path or environment that gets disturbed and restarting just initialises Keras settings.

picture sanjmeh on 10 Oct 2018
👍2

All 34 comments

My guess is that the problem here is that the GPU version of TensorFlow is
not able to load (likely b/c it can't load the CUDA and/or cuDNN
libraries). I'd start with installing the CPU version of TensorFlow and
confirming that it works, then getting back to the GPU version to see if
you can get it to work properly with CUDA, etc.

On Wednesday, August 16, 2017, Wade Cooper notifications@github.com wrote:

I got keras with tensorflow to install on R 3.4.1 via:

devtools::install_github("rstudio/keras")
library(keras)
system("conda config --set ssl_verify false") # get past conda sslerror's
install_keras(tensorflow = "gpu")

However now when I try to run the R sample code it's throwing Python
modules not found when attempting to load the mnist dataset:

library(keras)
mnist <- dataset_mnist()
Error in py_module_import(module, convert = convert) :
ModuleNotFoundError: No module named 'tensorflow'
In addition: Warning message:
In stop(e, call. = FALSE) : additional arguments ignored in stop()

So I loaded tensorflow and tried telling it the correct conda environment
manually but still gets errors for keras:

library(tensorflow)
use_condaenv('r-tensorflow')
mnist <- dataset_mnist()
Error: Python module keras was not found.

Detected Python configuration:

python: C:\Usersxyz\Desktop\work\software\Python\Conda3\python.exe
libpython: C:/Users/xyz/Desktop/work/software/Python/Conda3/python36.dll
pythonhome: C:\Usersxyz\Desktop\work\software\Python\Conda3
version: 3.6.0 |Anaconda 4.3.1 (64-bit)| (default, Dec 23 2016, 11:57:41) [MSC v.1900 64 bit (AMD64)]
Architecture: 64bit
numpy: C:\Usersxyz\Desktop\work\software\Python\Conda3\lib\site-packages\numpy
numpy_version: 1.11.3
keras: C:\Usersxyz\Desktop\work\software\Python\Conda3\lib\site-packages\keras

python versions found:
C:\Usersxyz\Desktop\work\software\Python\Conda3\python.exe
C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\env_full\python.exe
C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\r-tensorflow\python.exe
C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\tensorflow\python.exe

Note: I also have another conda environment set up running
keras/tensorflow fine via python code, not sure if there's some issues with
the use_condaenv() that it's not detecting correctly.

Thanks for any advice.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
https://github.com/rstudio/keras/issues/102, or mute the thread
https://github.com/notifications/unsubscribe-auth/AAGXx1RmgwawHLufq9cFvzm7WbZ0EYu6ks5sYwkngaJpZM4O5FIE
.

jjallaire picture jjallaire on 16 Aug 2017
👎2

I have the tensorflow GPU version running fine in Python, so CUDA is installed correctly and cuDNN is on my path.

I just did a fresh install, where first I 'conda remove --name r-tensorflow --all' to get rid of old env. But get the same issue:

library(keras)
system("conda config --set ssl_verify false")
# install_keras(tensorflow = "gpu")
install_keras()
mnist <- dataset_mnist()

This produces the same output:

> mnist <- dataset_mnist()
Error in py_module_import(module, convert = convert) : 
  ModuleNotFoundError: No module named 'tensorflow'
In addition: Warning message:
In stop(e, call. = FALSE) : additional arguments ignored in stop()
wtcooper picture wtcooper on 16 Aug 2017

It's possible though that it's not on your PATH within your R session (as
depending on what your R front end is it may be a desktop application like
R GUI or RStudio, in which case you may need to make special provision to
get the environment variables forwarded, see
https://tensorflow.rstudio.com/installation_gpu.html#prerequisties.

It may of course work fine from R in which case there is something else
going on, my suspicion comes from the fact that install_keras succeeded and
obviously it's finding keras (otherwise you'd get a "keras not found" error
message) but for some reason it can't find tensorflow. What is the output
of reticulate::py_config() after the error occurs?

Also, what is the output of the following in a fresh R session:

reticulate::py_discover_config("keras")
reticulate::py_discover_config("tensorflow")

On Wednesday, August 16, 2017, Wade Cooper notifications@github.com wrote:

I have the tensorflow GPU version running fine in Python, so CUDA is
installed correctly and cuDNN is on my path.

I just did a fresh install, where first I 'conda remove --name
r-tensorflow --all' to get rid of old env. But get the same issue:

library(keras)
system("conda config --set ssl_verify false")

install_keras(tensorflow = "gpu")

install_keras()
mnist <- dataset_mnist()

This produces the same output:

mnist <- dataset_mnist()
Error in py_module_import(module, convert = convert) :
ModuleNotFoundError: No module named 'tensorflow'
In addition: Warning message:
In stop(e, call. = FALSE) : additional arguments ignored in stop()

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/rstudio/keras/issues/102#issuecomment-322836434, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AAGXx1wCnUJyo-841VbFA5_7urglDUVvks5sYyDugaJpZM4O5FIE
.

jjallaire picture jjallaire on 16 Aug 2017

I have them showing up on my path in Sys.getenv():

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\libnvvp
C:\Users\xyz\Desktop\work\software\cuDNN\cuda\bin  (*v5.1)

Here's output from reticulate calls below in a fresh session. 

> reticulate::py_discover_config("keras")
python:         C:\Users\xyz\Desktop\work\software\Python\Conda3\python.exe
libpython:      C:/Users/xyz/Desktop/work/software/Python/Conda3/python36.dll
pythonhome:     C:\Users\xyz\Desktop\work\software\Python\Conda3
version:        3.6.0 |Anaconda 4.3.1 (64-bit)| (default, Dec 23 2016, 11:57:41) [MSC v.1900 64 bit (AMD64)]
Architecture:   64bit
numpy:          C:\Users\xyz\Desktop\work\software\Python\Conda3\lib\site-packages\numpy
numpy_version:  1.11.3
keras:          C:\Users\xyz\Desktop\work\software\Python\Conda3\lib\site-packages\keras

python versions found: 
 C:\Users\xyz\Desktop\work\software\Python\Conda3\python.exe
 C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\env_full\python.exe
 C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\r-tensorflow\python.exe
 C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\tensorflow\python.exe

> reticulate::py_discover_config("tensorflow")
python:         C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\r-tensorflow\python.exe
libpython:      C:/Users/xyz/Desktop/work/software/Python/Conda3/envs/r-tensorflow/python36.dll
pythonhome:     C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\R-TENS~1
version:        3.6.2 |Continuum Analytics, Inc.| (default, Jul 20 2017, 12:30:02) [MSC v.1900 64 bit (AMD64)]
Architecture:   64bit
numpy:          C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\R-TENS~1\lib\site-packages\numpy
numpy_version:  1.13.1
tensorflow:     C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\R-TENS~1\lib\site-packages\tensorflow

python versions found: 
 C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\r-tensorflow\python.exe
 C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\tensorflow\python.exe
 C:\Users\xyz\Desktop\work\software\Python\Conda3\python.exe
 C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\env_full\python.exe

Here's the output after the errors occur:

> library(keras)
> mnist <- dataset_mnist()
Error in py_module_import(module, convert = convert) : 
  ModuleNotFoundError: No module named 'tensorflow'
In addition: Warning message:
In stop(e, call. = FALSE) : additional arguments ignored in stop()
> reticulate::py_config()
python:         C:\Users\xyz\Desktop\work\software\Python\Conda3\python.exe
libpython:      C:/Users/xyz/Desktop/work/software/Python/Conda3/python36.dll
pythonhome:     C:\Users\xyz\Desktop\work\software\Python\Conda3
version:        3.6.0 |Anaconda 4.3.1 (64-bit)| (default, Dec 23 2016, 11:57:41) [MSC v.1900 64 bit (AMD64)]
Architecture:   64bit
numpy:          C:\Users\xyz\Desktop\work\software\Python\Conda3\lib\site-packages\numpy
numpy_version:  1.11.3
keras:          C:\Users\xyz\Desktop\work\software\Python\Conda3\lib\site-packages\keras

python versions found: 
 C:\Users\xyz\Desktop\work\software\Python\Conda3\python.exe
 C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\env_full\python.exe
 C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\r-tensorflow\python.exe
 C:\Users\xyz\Desktop\work\software\Python\Conda3\envs\tensorflow\python.exe
wtcooper picture wtcooper on 16 Aug 2017

I think I may see the problem. You have a version of keras installed in
your main Conda environment (the one at C:\Usersxyz\Desktop\work\
software\Python\Conda3\python.exe). However, that environment does not in
turn have tensorflow installed in it, and Keras wants to use tensorflow as
it's backend.

Alternatively, you may have tensorflow installed in the main conda
environment but it might for whatever reason not be loading.

A few things could resolve this:

1) Add a use_condaenv("r-tensorflow") right after library(keras) to force
it to use the conda env (bypassing the main conda environment which I
theorize has a problematic or non-existent tensorflow)

2) Install tensorflow within the main conda environment (you might have to
do that with the regular conda command line tools as install_keras wants to
use the r-tensorflow env)

3) Sys.setenv(KERAS_BACKEND="theano") to bypass tensorflow and just use
Theano as the back end.

On Wednesday, August 16, 2017, Wade Cooper notifications@github.com wrote:

I have them showing up on my path in Sys.getenv():

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\libnvvp
C:\Usersxyz\Desktop\work\software\cuDNN\cuda\bin (*v5.1)

Here's output from reticulate calls below in a fresh session.

reticulate::py_discover_config("keras")
python: C:\Usersxyz\Desktop\work\software\Python\Conda3\python.exe
libpython: C:/Users/xyz/Desktop/work/software/Python/Conda3/python36.dll
pythonhome: C:\Usersxyz\Desktop\work\software\Python\Conda3
version: 3.6.0 |Anaconda 4.3.1 (64-bit)| (default, Dec 23 2016, 11:57:41) [MSC v.1900 64 bit (AMD64)]
Architecture: 64bit
numpy: C:\Usersxyz\Desktop\work\software\Python\Conda3\lib\site-packages\numpy
numpy_version: 1.11.3
keras: C:\Usersxyz\Desktop\work\software\Python\Conda3\lib\site-packages\keras

python versions found:
C:\Usersxyz\Desktop\work\software\Python\Conda3\python.exe
C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\env_full\python.exe
C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\r-tensorflow\python.exe
C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\tensorflow\python.exe

reticulate::py_discover_config("tensorflow")
python: C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\r-tensorflow\python.exe
libpython: C:/Users/xyz/Desktop/work/software/Python/Conda3/envs/r-tensorflow/python36.dll
pythonhome: C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\R-TENS~1
version: 3.6.2 |Continuum Analytics, Inc.| (default, Jul 20 2017, 12:30:02) [MSC v.1900 64 bit (AMD64)]
Architecture: 64bit
numpy: C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\R-TENS~1\lib\site-packages\numpy
numpy_version: 1.13.1
tensorflow: C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\R-TENS~1\lib\site-packages\tensorflow

python versions found:
C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\r-tensorflow\python.exe
C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\tensorflow\python.exe
C:\Usersxyz\Desktop\work\software\Python\Conda3\python.exe
C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\env_full\python.exe

Here's the output after the errors occur:

library(keras)
mnist <- dataset_mnist()
Error in py_module_import(module, convert = convert) :
ModuleNotFoundError: No module named 'tensorflow'
In addition: Warning message:
In stop(e, call. = FALSE) : additional arguments ignored in stop()
reticulate::py_config()
python: C:\Usersxyz\Desktop\work\software\Python\Conda3\python.exe
libpython: C:/Users/xyz/Desktop/work/software/Python/Conda3/python36.dll
pythonhome: C:\Usersxyz\Desktop\work\software\Python\Conda3
version: 3.6.0 |Anaconda 4.3.1 (64-bit)| (default, Dec 23 2016, 11:57:41) [MSC v.1900 64 bit (AMD64)]
Architecture: 64bit
numpy: C:\Usersxyz\Desktop\work\software\Python\Conda3\lib\site-packages\numpy
numpy_version: 1.11.3
keras: C:\Usersxyz\Desktop\work\software\Python\Conda3\lib\site-packages\keras

python versions found:
C:\Usersxyz\Desktop\work\software\Python\Conda3\python.exe
C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\env_full\python.exe
C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\r-tensorflow\python.exe
C:\Usersxyz\Desktop\work\software\Python\Conda3\envs\tensorflow\python.exe

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/rstudio/keras/issues/102#issuecomment-322847275, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AAGXxy_hTtbVQWlTy66Rw3PzxNFp67gWks5sYyqsgaJpZM4O5FIE
.

jjallaire picture jjallaire on 16 Aug 2017

Yep that was it, not sure how I got keras in the base environment, must have forgotten to activate when I first installed it in another environment... I just deleted the keras folders in the base env and now it's downloading the mnist dataset fine in R.

I had tried to use:
use_condaenv("r-tensorflow")

But that apparently wasn't working, it was still defaulting to whatever was installed in the base conda environment. So maybe something is off with activating the correct environment. As long as I didn't have keras installed in the base environment, working like a champ now.

I just tried installing the gpu version and working fine too. I love the view_metrics, that's nice.

Also, not sure if helpful to add the ssl_verify snippet to the install documentation:
system("conda config --set ssl_verify false")
Might be others sitting behind firewalls that hit that issue and not sure how to deal with it.

Thanks!

wtcooper picture wtcooper on 16 Aug 2017

Okay, thanks for working this one through. I attempted to reproduce your exact scenario (keras in the base conda environment with no tensorflow and use_condaenv("r-tensorflow") and in my setup the base conda environment was successfully bypassed. I'll keep an eye on any other issue reports like this one to see if there is something we can do better.

jjallaire picture jjallaire on 17 Aug 2017

Hi, I seem to have a similar issue.

> mnist <- dataset_mnist() 
Error: Python module keras was not found.

I've uninstalled and reinstalled Keras and Tensorflow in both the main environment and the r-tensorflow environment.

If I install the cpu version of tensorflow in r-tensorflow, it seems to work. It is just the gpu version that cannot find keras. I found a self check script (https://gist.github.com/mrry/ee5dbcfdd045fa48a27d56664411d41c) and discovered that I needed to upgrade the cuDNN library. This seems to have fixed the issue.

Here is the output of tensorflow_self_check.py on my system before fixing the dll.

(r-tensorflow) C:\Users\jeff->python "tensorflow_self_check.py"
ERROR: Failed to import the TensorFlow module.

- Python version is 3.6.

- No module named TensorFlow is installed in this Python environment. You may
  install it using the command `pip install tensorflow`.

- Could not find cuDNN 6.

  The GPU version of TensorFlow requires that the correct cuDNN DLL be installed
  in a directory that is named in your %PATH% environment variable. Note that
  installing cuDNN is a separate step from installing CUDA, and it is often
  found in a different directory from the CUDA DLLs. The correct version of
  cuDNN depends on your version of TensorFlow:

  * TensorFlow 1.2.1 or earlier requires cuDNN 5.1. ('cudnn64_5.dll')
  * TensorFlow 1.3 or later requires cuDNN 6. ('cudnn64_6.dll')

  You may install the necessary DLL by downloading cuDNN from this URL:
  https://developer.nvidia.com/cudnn

After correcting the dll, I get notice of successful install.

(r-tensorflow) C:\Users\jeff->python "tensorflow_self_check.py"
TensorFlow successfully installed.
The installed version of TensorFlow includes GPU support.

For reference: I'm running Windows 10, RStudio 1.0.153, R x64 3.4.1

JeffHebert picture JeffHebert on 24 Sep 2017

Could you provide the output of:

library(reticulate)
py_discover_config('keras')
jjallaire picture jjallaire on 24 Sep 2017

Does the rstudio/keras actually use the python keras? I got similar problem, but can partially solve it by manually install keras through conda. Then it can detect the keras module.

But installing keras in conda will automatically install the "default" CPU version of tensorflow. Below is the output of "conda list" on "r-tensorflow", The 'tensorflow-gpu' is the GPU tensorflow that I installed manually, and the rest three are installed along with keras.

tensorflow                1.3.0                         0  
tensorflow-base           1.3.0                    py36_0  
tensorflow-gpu            1.1.0                     <pip>
tensorflow-tensorboard    0.1.5                    py36_0

The bad news is the "uninstall tensorflow and re-install tensorflow" trick won't work because conda remove the "default" tensorflow will automatically remove the keras too.

It will be very helpful if someone can help me understand the correct steps to install all the keras/tensorflow with GPU support and can be used in rstudio/keras. Plus some troubleshooting ...

Thanks!

zhukunism picture zhukunism on 29 Sep 2017

Yes it does use python keras.

The best guide to GPU installation is here: https://tensorflow.rstudio.com/tensorflow/articles/installation_gpu.html

I see that you are running v1.1 of TensorFlow for the GPU version. That may indicate that you are on OS X? If so then it can be a bit more difficult to get things working. In the case of all platforms problems with the GPU version are inevitably tied to not having the correct environment variables available (which results in the R session not being able to load CUDA and or cuDNN). Pay careful attention to all of the details in the article and you should be able to get it working!

jjallaire picture jjallaire on 29 Sep 2017

I am using Ubuntu.

I have installed tensorflow 1.2 on the same machine and running well with GPU. But got no luck when switching to R environment after trying 1.1,1.2,1.3 with GPU by running install_tensorflow("1.x-gpu'). So do I actually need to install python keras through conda or pip after github_install rstudio/keras? Or the rstudio/keras should automatically installed required python keras inside the default "r-tensorflow" env?

Also another question, tensorflow starts to have keras within its contrib package. And that's how I managed to run keras with GPU in python. Is it possible or any plan to have rstudio/keras running the "native" tensorflow/keras instead of the original keras?

Thanks!

zhukunism picture zhukunism on 29 Sep 2017

We have the option of either using the standalone keras Python package or using tf.contrib.keras within TensorFlow. The docs on this are here: https://keras.rstudio.com/articles/backend.html

From what I can tell you have a couple of different ways to go:

1) Use Sys.setenv(KERAS_IMPLEMENTATION = "tensorflow") to request the use of the version of Keras embedded in TensorFlow.

2) Install the keras python package within the environment you want to use. You can do this via keras::install_keras (which is just an alias for tensorflow::install_tensorflow that also installs the keras package alongside TensorFlow).

jjallaire picture jjallaire on 29 Sep 2017
👍1

I had a number of what look like PATH issues on Arch Linux using the "conda" install method. I switched to using the "virtualenv" method and everything works. Anaconda / Miniconda can do things to your PATH settings and I need to look at what it's doing; the symptom was that is_keras_available() returned FALSE.

znmeb picture znmeb on 29 Sep 2017

I had the same issue as the others in this thread. I'm on windows 10. Finally got it working. Went through the steps in the following general order:

  1. Installed anaconda
  2. library(keras) Didn't work with just install_keras() in R for some reason. I might have ran devtools::install_github("rstudio/keras") before anaconda install, don't remember exactly.
  3. Errors. Had to change main python path to the conda python with use_python().
  4. Installed keras and tensorflow in main conda env in the prompt.
  5. CPU version of keras/tensorflow now worked fine. Recognized keras.
  6. Installed CUDA 9.0 and cuDNN 6.
    6.5 Uninstalled CUDA 9.0 and installed CUDA 8.0.
  7. Re-tried install_tensorflow(version = "gpu") and then install_keras(tensorflow = "gpu") because I thought this was how I was supposed to install it. Realized I had two diffrent environments after installation was complete. Thought for a long time the initial main env installation was screwing with something (still hadn't run use_condaenv("r-tensorflow").
  8. Added PATH to 'cudnn64_6.dll' after errors when periodically using reticulate::import("keras.models") and other troubleshooting tips suggested here.
  9. Error: Python module keras was not found. After some googling and troubleshooting I tried use_condaenv("r-tensorflow") and later Sys.setenv(KERAS_IMPLEMENTATION = "tensorflow"). 
Detected Python configuration:

python:         C:\Users\Faton\Anaconda3\envs\r-tensorflow\python.exe
libpython:      C:/Users/Faton/Anaconda3/envs/r-tensorflow/python36.dll
pythonhome:     C:\Users\Faton\ANACON~1\envs\R-TENS~1
version:        3.6.3 |Anaconda, Inc.| (default, Nov  8 2017, 15:10:56) [MSC v.1900 64 bit (AMD64)]
Architecture:   64bit
numpy:          C:\Users\Faton\ANACON~1\envs\R-TENS~1\lib\site-packages\numpy
numpy_version:  1.13.3
keras:          C:\Users\Faton\ANACON~1\envs\R-TENS~1\lib\site-packages\keras

  1. Error: Python module keras was not found. Still getting the same error. Checked if keras is infact installed in the folder path: yes, it was. So I still go around thinking the order of installation screwed with something.

  2. Re-ran reticulate::import("keras.models"). It apparently still couldn't find 'cudnn64_6.dll'.

  3. Checked the path folder. I had downloaded and installed cudnn 7 for cuda 8.0...
  4. Downloaded cudnn6 for cuda 8.0 and unpacked in same folders.
  5. Error: Python module keras was not found. WHAT. THE. HELL?
  6. Chased some more dead ends.

  7. Changed PATH from C:\NVIDIA\cuda\bin to C:\NVIDIA\cuda\bin\. The former was suggested in the install instructions for cuDNN.

  8. Success.

In conclusion

It would have been immensely helpful if the error messages when attempting to run keras:: functions following a gpu-version install of tensorflow included more information than Error: Python module keras was not found. .

Please warn the user if the PATH to .dlls is not set correctly when attemping to run keras:: functions. I remember at some point I got some path errors (perhaps before I had installed CUDA). But I thought I had resolved them once the error messages stopped mentioning PATH and started saying Error: Python module keras was not found. .

It was an annoying issue, but mostly my own fault. Posting this message so other people running into something similar check their troubleshooting steps carefully.

Lauler picture Lauler on 2 Dec 2017

Yes agreed. In order to do that we need to introspect the system to try to sort out for the user what might be missing or misconfigured. This script looks like a promising start (https://gist.github.com/mrry/ee5dbcfdd045fa48a27d56664411d41c), we're hoping to incorporate this or something like it into the package soon.

jjallaire picture jjallaire on 3 Dec 2017

I am trying to install keras/tensorflow for Nvidia GPU under RStudio. The command:

install_keras(method="virtualenv",tensorflow="gpu")

Which culminates in a success message:
...
Collecting funcsigs>=1; python_version < "3.3" (from mock>=2.0.0->tensorflow-gpu)
Using cached funcsigs-1.0.2-py2.py3-none-any.whl
Collecting pbr>=0.11 (from mock>=2.0.0->tensorflow-gpu)
Using cached pbr-3.1.1-py2.py3-none-any.whl
Collecting setuptools (from protobuf>=3.2.0->tensorflow-gpu)
Using cached setuptools-38.2.5-py2.py3-none-any.whl
Installing collected packages: werkzeug, six, wheel, numpy, funcsigs, pbr, mock, setuptools, protobuf, tensorflow-gpu, h5py, pyyaml, idna, urllib3, certifi, chardet, requests, Pillow, scipy, keras
Successfully installed Pillow-5.0.0 certifi-2017.11.5 chardet-3.0.4 funcsigs-1.0.2 h5py-2.7.1 idna-2.6 keras-2.1.2 mock-2.0.0 numpy-1.13.3 pbr-3.1.1 protobuf-3.5.1 pyyaml-3.12 requests-2.18.4 scipy-1.0.0 setuptools-38.2.5 six-1.11.0 tensorflow-gpu-1.1.0 urllib3-1.22 werkzeug-0.14.1 wheel-0.30.0
Installation complete.
Restarting R session...

A find command shows that the virtualenv is created and that there is a tensorflow in site-packages:

find . -name tensorflow* -print
./.virtualenvs/r-tensorflow/lib/python2.7/site-packages/keras/backend/tensorflow_backend.py
./.virtualenvs/r-tensorflow/lib/python2.7/site-packages/keras/backend/tensorflow_backend.pyc
./.virtualenvs/r-tensorflow/lib/python2.7/site-packages/tensorflow
./.virtualenvs/r-tensorflow/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/dataframe/tensorflow_dataframe.py
./.virtualenvs/r-tensorflow/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/dataframe/tensorflow_dataframe.pyc
./.virtualenvs/r-tensorflow/lib/python2.7/site-packages/tensorflow/core/protobuf/tensorflow_server_pb2.py
./.virtualenvs/r-tensorflow/lib/python2.7/site-packages/tensorflow/core/protobuf/tensorflow_server_pb2.pyc
./.virtualenvs/r-tensorflow/lib/python2.7/site-packages/tensorflow/include/tensorflow
./.virtualenvs/r-tensorflow/lib/python2.7/site-packages/tensorflow/include/tensorflow/core/protobuf/tensorflow_server.pb.h
./.virtualenvs/r-tensorflow/lib/python2.7/site-packages/tensorflow_gpu-1.1.0.dist-info

Next reticulate commands seem to show the installations:

reticulate::py_discover_config("keras")
python: /Users/vince/.virtualenvs/r-tensorflow/bin/python
libpython: /System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/config/libpython2.7.dylib
pythonhome: /System/Library/Frameworks/Python.framework/Versions/2.7:/System/Library/Frameworks/Python.framework/Versions/2.7
virtualenv: /Users/vince/.virtualenvs/r-tensorflow/bin/activate_this.py
version: 2.7.10 (default, Feb 7 2017, 00:08:15) [GCC 4.2.1 Compatible Apple LLVM 8.0.0 (clang-800.0.34)]
numpy: /Users/vince/.virtualenvs/r-tensorflow/lib/python2.7/site-packages/numpy
numpy_version: 1.13.3
keras: /Users/vince/.virtualenvs/r-tensorflow/lib/python2.7/site-packages/keras

python versions found:
/Users/vince/.virtualenvs/r-tensorflow/bin/python
/usr/bin/python

reticulate::py_discover_config("tensorflow")
python: /Users/vince/.virtualenvs/r-tensorflow/bin/python
libpython: /System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/config/libpython2.7.dylib
pythonhome: /System/Library/Frameworks/Python.framework/Versions/2.7:/System/Library/Frameworks/Python.framework/Versions/2.7
virtualenv: /Users/vince/.virtualenvs/r-tensorflow/bin/activate_this.py
version: 2.7.10 (default, Feb 7 2017, 00:08:15) [GCC 4.2.1 Compatible Apple LLVM 8.0.0 (clang-800.0.34)]
numpy: /Users/vince/.virtualenvs/r-tensorflow/lib/python2.7/site-packages/numpy
numpy_version: 1.13.3
tensorflow: /Users/vince/.virtualenvs/r-tensorflow/lib/python2.7/site-packages/tensorflow

python versions found:
/Users/vince/.virtualenvs/r-tensorflow/bin/python
/usr/bin/python

But it won't find tensorflow:

library(tensorflow)
library(keras)
sess<-tf$session()
Error: Python module tensorflow was not found.

Detected Python configuration:

python: /Users/vince/.virtualenvs/r-tensorflow/bin/python
libpython: /System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/config/libpython2.7.dylib
pythonhome: /System/Library/Frameworks/Python.framework/Versions/2.7:/System/Library/Frameworks/Python.framework/Versions/2.7
virtualenv: /Users/vince/.virtualenvs/r-tensorflow/bin/activate_this.py
version: 2.7.10 (default, Feb 7 2017, 00:08:15) [GCC 4.2.1 Compatible Apple LLVM 8.0.0 (clang-800.0.34)]
numpy: /Users/vince/.virtualenvs/r-tensorflow/lib/python2.7/site-packages/numpy
numpy_version: 1.13.3
keras: /Users/vince/.virtualenvs/r-tensorflow/lib/python2.7/site-packages/keras

python versions found:
/Users/vince/.virtualenvs/r-tensorflow/bin/python
/usr/bin/python

Any help will be greatly appreciated.

vmstanford picture vmstanford on 2 Jan 2018

This error is nearly 100% of the time due to not having CUDA available on your LD_LIBRARY_PATH. Please review all of this documentation carefully: https://tensorflow.rstudio.com/tools/installation_gpu.html

jjallaire picture jjallaire on 2 Jan 2018

jjallaire,

Thank you for responding. I think I have a finer grained problem than that. My .profile file has the DYLD path set:

# MacPorts Installer addition on 2015-12-04_at_21:13:09: adding an appropriate PATH variable for use with MacPorts.                                                                                               
export PATH="/opt/local/bin:/opt/local/sbin:$PATH"

# Finished adapting your PATH environment variable for use with MacPorts.                                                                                                                                         

# added by Anaconda3 2.4.0 installer                                                                                                                                                                              
export PATH="/Users/vince/anaconda/bin:$PATH"                                                                                                                                                                   

# add ~/bin to path                                                                                                                                                                                               
export PATH="~/bin:$PATH"

# Nvidia and tensorflow stuff                                                                                                                                                                                     
export PATH="/usr/local/cuda/bin:$PATH"
export PATH="~.virtualenvs/r-tensorflow/bin/:$PATH"

# CUDA/cuDNN stuff                                                                                                                                                                                                
export  DYLD_LIBRARY_PATH=/usr/local/cuda/lib:$DYLD_LIBRARY_PATH

Moreover, I see cudnn in the lib as follows:

$ source .profile
$ ls /usr/local/cuda/lib

libcublas.9.0.dylib     libcusolver_static.a        libnppim.dylib
libcublas.dylib         libcusparse.9.0.dylib       libnppim_static.a
libcublas_device.a      libcusparse.dylib       libnppist.9.0.dylib
libcublas_static.a      libcusparse_static.a        libnppist.dylib
libcuda.dylib           libnppc.9.0.dylib       libnppist_static.a
libcudadevrt.a          libnppc.dylib           libnppisu.9.0.dylib
libcudart.9.0.dylib     libnppc_static.a        libnppisu.dylib
libcudart.dylib         libnppial.9.0.dylib     libnppisu_static.a
libcudart_static.a      libnppial.dylib         libnppitc.9.0.dylib
libcudnn.7.dylib        libnppial_static.a      libnppitc.dylib
libcudnn.dylib          libnppicc.9.0.dylib     libnppitc_static.a
libcudnn_static.a       libnppicc.dylib         libnpps.9.0.dylib
libcufft.9.0.dylib      libnppicc_static.a      libnpps.dylib
libcufft.dylib          libnppicom.9.0.dylib        libnpps_static.a
libcufft_static.a       libnppicom.dylib        libnvToolsExt.1.dylib
libcufftw.9.0.dylib     libnppicom_static.a     libnvToolsExt.dylib
libcufftw.dylib         libnppidei.9.0.dylib        libnvblas.9.0.dylib
libcufftw_static.a      libnppidei.dylib        libnvblas.dylib
libcuinj.9.0.dylib      libnppidei_static.a     libnvgraph.9.0.dylib
libcuinj.dylib          libnppif.9.0.dylib      libnvgraph.dylib
libculibos.a            libnppif.dylib          libnvgraph_static.a
libcurand.9.0.dylib     libnppif_static.a       libnvrtc-builtins.9.0.dylib
libcurand.dylib     libnppig.9.0.dylib      libnvrtc-builtins.dylib
libcurand_static.a      libnppig.dylib          libnvrtc.9.0.dylib
libcusolver.9.0.dylib   libnppig_static.a       libnvrtc.dylib
libcusolver.dylib       libnppim.9.0.dylib      stubs

I compiled and ran all the cuda demos in /usr/local/cuda/samples and they work fine.

Seems like some version problem. I am using:

xcode 8.3.3 (8E3004b)
cuda_9.0.176_mac.dmg
cudnn-9.0-osx-x64-v7.tgz
clang-8.02.0.42

HW: Mac Pro 5.1 with dual 6-core Xeon Westmere 2.93GHZ processors
Nvidia 970 4GB GPU

Any of these things incompatable?

Any other thoughts?

vmstanford picture vmstanford on 4 Jan 2018

If you are the Mac it's important to note that GPU support was dropped from the Mac version after TensorFlow v1.1, see https://tensorflow.rstudio.com/tools/installation_gpu.html#mac-os-x

jjallaire picture jjallaire on 4 Jan 2018

Thank you for the idea. I'll try the 1.1 tensorflow. There is an install
flag for that in the tensorflow and keras installers for R.

On Thu, Jan 4, 2018 at 3:44 PM, JJ Allaire notifications@github.com wrote:

If you are the Mac it's important to note that GPU support was dropped
from the Mac version after TensorFlow v1.1, see
https://tensorflow.rstudio.com/tools/installation_gpu.html#mac-os-x

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/rstudio/keras/issues/102#issuecomment-355393555, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AhZJksOO4cqYIIo-giC1K9X7tI83Gk19ks5tHTg9gaJpZM4O5FIE
.

vmstanford picture vmstanford on 4 Jan 2018

Hi JJ,

I just tried installing tensorflow 1.1.0 under keras. The problem
persists. keras says it is installing tf 1.1.0 but when I try tf hello
world it errors out as follows:

library(keras)
install_keras(method="virtualenv",tensorflow="1.1.0-gpu")
Using existing virtualenv at ~/.virtualenvs/r-tensorflow
Upgrading pip ...
Collecting pip
Using cached pip-9.0.1-py2.py3-none-any.whl
Installing collected packages: pip
Successfully installed pip-9.0.1
Upgrading wheel ...
Collecting wheel
Using cached wheel-0.30.0-py2.py3-none-any.whl
Installing collected packages: wheel
Successfully installed wheel-0.30.0
Upgrading setuptools ...
Collecting setuptools
Using cached setuptools-38.2.5-py2.py3-none-any.whl
Installing collected packages: setuptools
Successfully installed setuptools-38.2.5
Installing TensorFlow ...
Collecting tensorflow-gpu==1.1.0
Using cached tensorflow_gpu-1.1.0-cp27-cp27m-macosx_10_11_x86_64.whl
Collecting h5py
Using cached
h5py-2.7.1-cp27-cp27m-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl
Collecting pyyaml
Collecting requests
Using cached requests-2.18.4-py2.py3-none-any.whl
Collecting Pillow
Using cached
Pillow-5.0.0-cp27-cp27m-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl
Collecting keras
Using cached Keras-2.1.2-py2.py3-none-any.whl
Collecting scipy
Using cached
scipy-1.0.0-cp27-cp27m-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl
Collecting werkzeug>=0.11.10 (from tensorflow-gpu==1.1.0)
Using cached Werkzeug-0.14.1-py2.py3-none-any.whl
Collecting six>=1.10.0 (from tensorflow-gpu==1.1.0)
Using cached six-1.11.0-py2.py3-none-any.whl
Collecting wheel (from tensorflow-gpu==1.1.0)
Using cached wheel-0.30.0-py2.py3-none-any.whl
Collecting numpy>=1.11.0 (from tensorflow-gpu==1.1.0)
Using cached
numpy-1.13.3-cp27-cp27m-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl
Collecting mock>=2.0.0 (from tensorflow-gpu==1.1.0)
Using cached mock-2.0.0-py2.py3-none-any.whl
Collecting protobuf>=3.2.0 (from tensorflow-gpu==1.1.0)
Using cached protobuf-3.5.1-py2.py3-none-any.whl
Collecting idna<2.7,>=2.5 (from requests)
Using cached idna-2.6-py2.py3-none-any.whl
Collecting urllib3<1.23,>=1.21.1 (from requests)
Using cached urllib3-1.22-py2.py3-none-any.whl
Collecting certifi>=2017.4.17 (from requests)
Using cached certifi-2017.11.5-py2.py3-none-any.whl
Collecting chardet<3.1.0,>=3.0.2 (from requests)
Using cached chardet-3.0.4-py2.py3-none-any.whl
Collecting funcsigs>=1; python_version < "3.3" (from
mock>=2.0.0->tensorflow-gpu==1.1.0)
Using cached funcsigs-1.0.2-py2.py3-none-any.whl
Collecting pbr>=0.11 (from mock>=2.0.0->tensorflow-gpu==1.1.0)
Using cached pbr-3.1.1-py2.py3-none-any.whl
Collecting setuptools (from protobuf>=3.2.0->tensorflow-gpu==1.1.0)
Using cached setuptools-38.2.5-py2.py3-none-any.whl
Installing collected packages: werkzeug, six, wheel, numpy, funcsigs, pbr,
mock, setuptools, protobuf, tensorflow-gpu, h5py, pyyaml, idna, urllib3,
certifi, chardet, requests, Pillow, scipy, keras
Successfully installed Pillow-5.0.0 certifi-2017.11.5 chardet-3.0.4
funcsigs-1.0.2 h5py-2.7.1 idna-2.6 keras-2.1.2 mock-2.0.0 numpy-1.13.3
pbr-3.1.1 protobuf-3.5.1 pyyaml-3.12 requests-2.18.4 scipy-1.0.0
setuptools-38.2.5 six-1.11.0 tensorflow-gpu-1.1.0 urllib3-1.22
werkzeug-0.14.1 wheel-0.30.0

Installation complete.

Restarting R session...

library(keras)
library(tensorflow)
sess<-tf$Session()
Error: Python module tensorflow was not found.

Detected Python configuration:

python: /Users/vince/.virtualenvs/r-tensorflow/bin/python
libpython:
/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/config/libpython2.7.dylib
pythonhome:
/System/Library/Frameworks/Python.framework/Versions/2.7:/System/Library/Frameworks/Python.framework/Versions/2.7
virtualenv: /Users/vince/.virtualenvs/r-tensorflow/bin/activate_this.py
version: 2.7.10 (default, Feb 7 2017, 00:08:15) [GCC 4.2.1
Compatible Apple LLVM 8.0.0 (clang-800.0.34)]
numpy:
/Users/vince/.virtualenvs/r-tensorflow/lib/python2.7/site-packages/numpy
numpy_version: 1.13.3
keras:
/Users/vince/.virtualenvs/r-tensorflow/lib/python2.7/site-packages/keras

python versions found:
/Users/vince/.virtualenvs/r-tensorflow/bin/python
/usr/bin/python

On Thu, Jan 4, 2018 at 3:44 PM, JJ Allaire notifications@github.com wrote:

If you are the Mac it's important to note that GPU support was dropped
from the Mac version after TensorFlow v1.1, see
https://tensorflow.rstudio.com/tools/installation_gpu.html#mac-os-x

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/rstudio/keras/issues/102#issuecomment-355393555, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AhZJksOO4cqYIIo-giC1K9X7tI83Gk19ks5tHTg9gaJpZM4O5FIE
.

vmstanford picture vmstanford on 4 Jan 2018

It must not be able to find your CUDA libraries. Everything I know about this subject is here: https://tensorflow.rstudio.com/tools/installation_gpu.html#mac-os-x. Note that if you start RStudio from the dock it won't see .bash_profile.

jjallaire picture jjallaire on 5 Jan 2018

I usually do start RStudio from the dock, so that could be a problem...

I just ran the install from command line R, and the symptoms are the same.
Emmm...

Perhaps I should try the CPU version again. Ran tf hello world fine with
that. Got to figure out how to link in MKL. With twelve cores I should be
able to get some speed up that way.

On Thu, Jan 4, 2018 at 6:44 PM, JJ Allaire notifications@github.com wrote:

It must not be able to find your CUDA libraries. Everything I know about
this subject is here: https://tensorflow.rstudio.
com/tools/installation_gpu.html#mac-os-x. Note that if you start RStudio
from the dock it won't see .bash_profile.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/rstudio/keras/issues/102#issuecomment-355431865, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AhZJkqwrcAGQNcoENv_ziHJXKTkqqrTiks5tHWJ0gaJpZM4O5FIE
.

vmstanford picture vmstanford on 5 Jan 2018

I was facing issue of seeing error Python module keras was not found

When I tried :

library(reticulate)
py_discover_config('keras')

The output message showed keras : keras not found.

So following this thread, somewhere up it was mentioned to try keras::install_keras() in a fresh session. I did that and noticed that it was infact the keras python package that it needed to install.

This was the output of it:

> keras::install_keras()
Using r-tensorflow conda environment for TensorFlow installation
Fetching package metadata .............
Solving package specifications: .

Package plan for installation in environment C:\ANACON~1\envs\r-tensorflow:

The following NEW packages will be INSTALLED:

    asn1crypto:      0.22.0-py36_0     conda-forge
    bleach:          1.5.0-py36_0      conda-forge
    ca-certificates: 2017.11.5-0       conda-forge
    certifi:         2017.11.5-py36_0  conda-forge
    cffi:            1.11.2-py36_0     conda-forge
    chardet:         3.0.4-py36_0      conda-forge
    cryptography:    2.1.4-py36_0      conda-forge
    html5lib:        0.9999999-py36_0  conda-forge
    idna:            2.6-py36_1        conda-forge
    keras:           2.0.9-py36_0      conda-forge
    libgpuarray:     0.7.5-vc14_0      conda-forge [vc14]
    mako:            1.0.7-py36_0      conda-forge
    markdown:        2.6.9-py36_0      conda-forge
    markupsafe:      1.0-py36_0        conda-forge
    openssl:         1.0.2n-vc14_0     conda-forge [vc14]
    protobuf:        3.5.1-py36_vc14_2 conda-forge [vc14]
    pycparser:       2.18-py36_0       conda-forge
    pygpu:           0.7.5-py36_0      conda-forge
    pyopenssl:       17.2.0-py36_0     conda-forge
    pysocks:         1.6.7-py36_0      conda-forge
    tensorboard:     0.4.0rc3-py36_2   conda-forge
    tensorflow:      1.4.0-py36_0      conda-forge
    theano:          0.9.0-py36_1      conda-forge
    urllib3:         1.22-py36_0       conda-forge
    webencodings:    0.5-py36_0        conda-forge
    werkzeug:        0.12.2-py_1       conda-forge
    win_inet_pton:   1.0.1-py36_1      conda-forge
    yaml:            0.1.7-vc14_0      conda-forge [vc14]

The following packages will be UPDATED:

    freetype:        2.5.5-vc14_2                  --> 2.8.1-vc14_0  conda-forge [vc14]
    h5py:            2.7.0-np113py36_0             --> 2.7.1-py36_2  conda-forge
    hdf5:            1.8.15.1-vc14_4               --> 1.10.1-vc14_1 conda-forge [vc14]
    jpeg:            9b-vc14_0                     --> 9b-vc14_2     conda-forge [vc14]
    libpng:          1.6.27-vc14_0                 --> 1.6.34-vc14_0 conda-forge [vc14]
    libtiff:         4.0.6-vc14_3                  --> 4.0.9-vc14_0  conda-forge [vc14]
    pillow:          4.2.1-py36_0                  --> 4.3.0-py36_2  conda-forge
    pyyaml:          3.12-py36_0                   --> 3.12-py36_1   conda-forge
    requests:        2.14.2-py36_0                 --> 2.18.4-py36_1 conda-forge
    zlib:            1.2.8-vc14_3                  --> 1.2.11-vc14_0 conda-forge [vc14]

The following packages will be SUPERCEDED by a higher-priority channel:

    bzip2:           1.0.6-vc14_3                  --> 1.0.6-vc14_1  conda-forge [vc14]

ca-certificate 100% |###############################| Time: 0:00:00 501.42 kB/s
bzip2-1.0.6-vc 100% |###############################| Time: 0:00:00 561.50 kB/s
libgpuarray-0. 100% |###############################| Time: 0:00:00 866.81 kB/s
openssl-1.0.2n 100% |###############################| Time: 0:00:02   2.08 MB/s
yaml-0.1.7-vc1 100% |###############################| Time: 0:00:00 452.63 kB/s
zlib-1.2.11-vc 100% |###############################| Time: 0:00:00 727.69 kB/s
hdf5-1.10.1-vc 100% |###############################| Time: 0:00:05   3.46 MB/s
jpeg-9b-vc14_2 100% |###############################| Time: 0:00:00   2.62 MB/s
libpng-1.6.34- 100% |###############################| Time: 0:00:00   1.75 MB/s
asn1crypto-0.2 100% |###############################| Time: 0:00:00   1.56 MB/s
certifi-2017.1 100% |###############################| Time: 0:00:00 906.56 kB/s
chardet-3.0.4- 100% |###############################| Time: 0:00:00 655.77 kB/s
freetype-2.8.1 100% |###############################| Time: 0:00:00   1.19 MB/s
idna-2.6-py36_ 100% |###############################| Time: 0:00:00   1.02 MB/s
libtiff-4.0.9- 100% |###############################| Time: 0:00:00 764.03 kB/s
markdown-2.6.9 100% |###############################| Time: 0:00:00 899.24 kB/s
markupsafe-1.0 100% |###############################| Time: 0:00:00   1.08 MB/s
pycparser-2.18 100% |###############################| Time: 0:00:00   1.26 MB/s
pyyaml-3.12-py 100% |###############################| Time: 0:00:00 775.30 kB/s
webencodings-0 100% |###############################| Time: 0:00:00   1.10 MB/s
werkzeug-0.12. 100% |###############################| Time: 0:00:00 728.63 kB/s
win_inet_pton- 100% |###############################| Time: 0:00:00 306.80 kB/s
cffi-1.11.2-py 100% |###############################| Time: 0:00:00 615.77 kB/s
h5py-2.7.1-py3 100% |###############################| Time: 0:00:00   1.27 MB/s
html5lib-0.999 100% |###############################| Time: 0:00:00   1.10 MB/s
mako-1.0.7-py3 100% |###############################| Time: 0:00:00 449.23 kB/s
pillow-4.3.0-p 100% |###############################| Time: 0:00:00   1.49 MB/s
protobuf-3.5.1 100% |###############################| Time: 0:00:03   3.25 MB/s
pysocks-1.6.7- 100% |###############################| Time: 0:00:00   1.28 MB/s
bleach-1.5.0-p 100% |###############################| Time: 0:00:00 427.06 kB/s
cryptography-2 100% |###############################| Time: 0:00:00   4.48 MB/s
pygpu-0.7.5-py 100% |###############################| Time: 0:00:00   3.20 MB/s
pyopenssl-17.2 100% |###############################| Time: 0:00:00   4.58 MB/s
tensorboard-0. 100% |###############################| Time: 0:00:00   3.41 MB/s
theano-0.9.0-p 100% |###############################| Time: 0:00:01   3.62 MB/s
keras-2.0.9-py 100% |###############################| Time: 0:00:01 299.70 kB/s
tensorflow-1.4 100% |###############################| Time: 0:00:07   3.24 MB/s
urllib3-1.22-p 100% |###############################| Time: 0:00:00   2.00 MB/s
requests-2.18. 100% |###############################| Time: 0:00:00 353.38 kB/s

Installation complete.

Once done, I tried to run my code again and it worked.

Also the py_discover_config('keras') showed keras now pointing to keras python package in the tensorflow env directory.

python:         C:\ANACON~1\envs\R-TENS~1\python.exe
libpython:      C:/ANACON~1/envs/R-TENS~1/python36.dll
pythonhome:     C:\ANACON~1\envs\R-TENS~1
version:        3.6.2 |Continuum Analytics, Inc.| (default, Jul 20 2017, 12:30:02) [MSC v.1900 64 bit (AMD64)]
Architecture:   64bit
numpy:          C:\ANACON~1\envs\R-TENS~1\lib\site-packages\numpy
numpy_version:  1.13.3
keras:          C:\ANACON~1\envs\R-TENS~1\lib\site-packages\keras

python versions found: 
 C:\ANACON~1\envs\R-TENS~1\python.exe
 C:\ANACON~1\python.exe
 C:\Anaconda3\envs\py27\python.exe
 C:\Anaconda3\envs\r-tensorflow\python.exe

Hope this helps someone facing the same issue like me and resolve it quicker.

tarunparmar picture tarunparmar on 5 Jan 2018

I had the same problem, which I guess the cause is having multiple python versions.

installing missing packages by running pip from run console (CTRL+R)
pip install keras
pip install tensorflow

check if any package missing with these

reticulate::py_discover_config("keras")
reticulate::py_discover_config("tensorflow")
selcukakbas picture selcukakbas on 5 Feb 2018

Finally got tensorflow and keras with GPU up running on rstudio. Some tips learned are here.
I am using Ubuntu, and rstudio server, installed tensorflow 1.5 from https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.5.0-cp36-cp36m-linux_x86_64.whl

  1. You can check tensorflow and keras python module healthiness by these 2 commands
py_module_available('keras')
py_module_available('tensorflow')

make sure they all return TRUE, otherwise you will see this "Python module not found error"

  1. You must install CUDA 9.0 (for tensorflow 1.5), not 8.0, not 9.1. Tensorflow needs very specific version (at least from my experience) to work with GPU. Incorrect cuda version can cause the "Python module not found error". This https://tensorflow.rstudio.com/tensorflow/articles/installation_gpu.html is very useful on guiding you thru the process.

  2. Install rstudio keras package by running 

install.packages("keras")

  1. Install all the python modules. If you are using rstudio server, make sure you run you installation in a R console rather than browser thru server. It caused me some trouble as the server might fail to read correct env var as they read them thru slightly different user profile. But once all things installed, you can run keras in browser without any trouble.

    There are three steps.
    i. create conda "r-tensorflow" env
    ii. install tensorflow 1.5 with gpu
    iii. install keras
    For all of these 3 steps, you can either do it manually using conda by yourself, or by one command in R:

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

However I recommand you always try to install them in R other than thru conda manually. Esp. for keras installation. Somehow in my case, the numpy comes with the keras is not compatible with tensorflow when running in R. So what I did is manually create the "r-tensorflow" in conda, and install tensorflow inside conda. but then running following command in R to let R-keras pick up the correct model to install:

tensorflow::install_tensorflow_extras("keras")
  1. If all things done, you should be able to run these lines without any error:
library(keras)
use_condaenv("r-tensorflow",required=T)
data <- dataset_mnist()
zhukunism picture zhukunism on 11 Feb 2018

Hello zhukunism - are you able to provide step by step instructions? I have just about tried every and any possible combinations and cant make it work on my Windows 10/ RStudio machine

KKel414 picture KKel414 on 14 Feb 2018

@KKel414 , i am using Ubuntu with Rstudio server and code from browser. Assuming tf v1.5 with GPU.

  1. Install CUDA v9.0. strictly follow the steps from this https://tensorflow.rstudio.com/tensorflow/articles/installation_gpu.html
  2. Install anaconda. I am using a older version of 4.3.14 with python 3.6. you can the latest 5.1, but not sure if works.
  3. Install the R-keras package (provided by rstudio, not the python keras) by running
install.packages("keras")
  1. Running this command to (i) create r-tensorflow conda env, (ii) install tf v1.5, (iii) install keras (the actual keras module)
library(keras)
keras::install_keras(tensorflow = "gpu",method="conda")
  1. If all installation successful, you should be able to run following command without trouble
library(keras)
use_condaenv("r-tensorflow",required=T)
data <- dataset_mnist()

if not, below commands are used for do some troubleshooting:

py_module_available('keras') # must return TRUE
py_module_available('tensorflow') # must return TRUE
py_discover_config("keras") # more info on the python env, tf and keras

Good luck.

zhukunism picture zhukunism on 22 Feb 2018

This issue seems to be reappearing from time to time every month on the same machine & same Rstudio. I have started dreading this error "Python module not found". Every time I forget what was the solution and I start to search and land on this GitHub thread. Waste a lot of time reading the entire thread eliminating each possibility - painfully & well explained.
But at the end, just a restart of Rstudio works.

I think there is some path or environment that gets disturbed and restarting just initialises Keras settings.

sanjmeh picture sanjmeh on 10 Oct 2018
👍2

Hi, all
I think I have the same issue
```{r}

library(keras)
library(tensorflow)
minst <- dataset_mnist()
Error: Python module keras was not found.

Detected Python configuration:

python: /usr/bin/python
libpython: /System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/config/libpython2.7.dylib
pythonhome: /System/Library/Frameworks/Python.framework/Versions/2.7:/System/Library/Frameworks/Python.framework/Versions/2.7
version: 2.7.10 (default, Jul 15 2017, 17:16:57) [GCC 4.2.1 Compatible Apple LLVM 9.0.0 (clang-900.0.31)]
numpy: /System/Library/Frameworks/Python.framework/Versions/2.7/Extras/lib/python/numpy
numpy_version: 1.8.0
keras: [NOT FOUND]

python versions found:
/anaconda3/envs/r-tensorflow/bin/python
/usr/bin/python

and the following code showed
```{r}
install_keras()

Remove all packages in environment /anaconda3/envs/r-tensorflow:


## Package Plan ##

  environment location: /anaconda3/envs/r-tensorflow


The following packages will be REMOVED:

    asn1crypto:      0.24.0-py36_1003                       conda-forge
    blas:            1.1-openblas                           conda-forge
    ca-certificates: 2018.11.29-ha4d7672_0                  conda-forge
    certifi:         2018.11.29-py36_1000                   conda-forge
    cffi:            1.11.5-py36h342bebf_1001               conda-forge
    chardet:         3.0.4-py36_1003                        conda-forge
    cryptography:    2.3.1-py36hdbc3d79_1000                conda-forge
    freetype:        2.9.1-h597ad8a_1005                    conda-forge
    h5py:            2.9.0-nompi_py36h35dc34b_1102          conda-forge
    hdf5:            1.10.4-nompi_h646315f_1105             conda-forge
    idna:            2.8-py36_1000                          conda-forge
    jpeg:            9c-h1de35cc_1001                       conda-forge
    libcxx:          4.0.1-hcfea43d_1                                  
    libcxxabi:       4.0.1-hcfea43d_1                                  
    libedit:         3.1.20181209-hb402a30_0                           
    libffi:          3.2.1-h475c297_4                                  
    libgfortran:     3.0.1-h93005f0_2                                  
    libpng:          1.6.36-ha441bb4_1000                   conda-forge
    libtiff:         4.0.10-h79f4b77_1001                   conda-forge
    ncurses:         6.1-h0a44026_1                                    
    numpy:           1.16.0-py36_blas_openblash4251c03_1000 conda-forge [blas_openblas]
    olefile:         0.46-py_0                              conda-forge
    openblas:        0.3.3-hdc02c5d_1001                    conda-forge
    openssl:         1.0.2p-h1de35cc_1002                   conda-forge
    pillow:          5.4.1-py36hbddbef0_1000                conda-forge
    pip:             18.1-py36_0                                       
    pycparser:       2.19-py_0                              conda-forge
    pyopenssl:       18.0.0-py36_1000                       conda-forge
    pysocks:         1.6.8-py36_1002                        conda-forge
    python:          3.6.7-h4a56312_1001                    conda-forge
    pyyaml:          3.13-py36h1de35cc_1001                 conda-forge
    readline:        7.0-h1de35cc_5                                    
    requests:        2.21.0-py36_1000                       conda-forge
    scipy:           1.2.0-py36_blas_openblash486cb9f_1201  conda-forge [blas_openblas]
    setuptools:      40.6.3-py36_0                                     
    six:             1.12.0-py36_1000                       conda-forge
    sqlite:          3.26.0-ha441bb4_0                                 
    tk:              8.6.9-ha441bb4_1000                    conda-forge
    urllib3:         1.24.1-py36_1000                       conda-forge
    wheel:           0.32.3-py36_0                                     
    xz:              5.2.4-h1de35cc_4                                  
    yaml:            0.1.7-h1de35cc_1001                    conda-forge
    zlib:            1.2.11-h1de35cc_3                                 

Creating r-tensorflow conda environment for TensorFlow installation...
Solving environment: ...working... done

## Package Plan ##

  environment location: /anaconda3/envs/r-tensorflow

  added / updated specs: 
    - python=3.6


The following NEW packages will be INSTALLED:

    ca-certificates: 2018.12.5-0            
    certifi:         2018.11.29-py36_0      
    libcxx:          4.0.1-hcfea43d_1       
    libcxxabi:       4.0.1-hcfea43d_1       
    libedit:         3.1.20181209-hb402a30_0
    libffi:          3.2.1-h475c297_4       
    ncurses:         6.1-h0a44026_1         
    openssl:         1.1.1a-h1de35cc_0      
    pip:             18.1-py36_0            
    python:          3.6.8-haf84260_0       
    readline:        7.0-h1de35cc_5         
    setuptools:      40.6.3-py36_0          
    sqlite:          3.26.0-ha441bb4_0      
    tk:              8.6.8-ha441bb4_0       
    wheel:           0.32.3-py36_0          
    xz:              5.2.4-h1de35cc_4       
    zlib:            1.2.11-h1de35cc_3      

Preparing transaction: ...working... done
Verifying transaction: ...working... done
Executing transaction: ...working... done
#
# To activate this environment, use:
# > source activate r-tensorflow
#
# To deactivate an active environment, use:
# > source deactivate
#



==> WARNING: A newer version of conda exists. <==
  current version: 4.4.10
  latest version: 4.6.1

Please update conda by running

    $ conda update -n base conda


Installing TensorFlow...
Collecting tensorflow==1.10.0 from https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-1.10.0-py3-none-any.whl
  Using cached https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-1.10.0-py3-none-any.whl
Collecting setuptools<=39.1.0 (from tensorflow==1.10.0)
  Using cached https://files.pythonhosted.org/packages/8c/10/79282747f9169f21c053c562a0baa21815a8c7879be97abd930dbcf862e8/setuptools-39.1.0-py2.py3-none-any.whl
Collecting termcolor>=1.1.0 (from tensorflow==1.10.0)
Collecting absl-py>=0.1.6 (from tensorflow==1.10.0)
Collecting tensorboard<1.11.0,>=1.10.0 (from tensorflow==1.10.0)
  Using cached https://files.pythonhosted.org/packages/c6/17/ecd918a004f297955c30b4fffbea100b1606c225dbf0443264012773c3ff/tensorboard-1.10.0-py3-none-any.whl
Collecting numpy<=1.14.5,>=1.13.3 (from tensorflow==1.10.0)
  Using cached https://files.pythonhosted.org/packages/f6/cd/b2c50b5190b66c711c23ef23c41d450297eb5a54d2033f8dcb3b8b13ac85/numpy-1.14.5-cp36-cp36m-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl
Collecting astor>=0.6.0 (from tensorflow==1.10.0)
  Using cached https://files.pythonhosted.org/packages/35/6b/11530768cac581a12952a2aad00e1526b89d242d0b9f59534ef6e6a1752f/astor-0.7.1-py2.py3-none-any.whl
Collecting wheel>=0.26 (from tensorflow==1.10.0)
  Using cached https://files.pythonhosted.org/packages/ff/47/1dfa4795e24fd6f93d5d58602dd716c3f101cfd5a77cd9acbe519b44a0a9/wheel-0.32.3-py2.py3-none-any.whl
Collecting six>=1.10.0 (from tensorflow==1.10.0)
  Using cached https://files.pythonhosted.org/packages/73/fb/00a976f728d0d1fecfe898238ce23f502a721c0ac0ecfedb80e0d88c64e9/six-1.12.0-py2.py3-none-any.whl
Collecting protobuf>=3.6.0 (from tensorflow==1.10.0)
  Using cached https://files.pythonhosted.org/packages/c7/27/133f225035b9539f2dcfebcdf9a69ff0152f56e0120160ec5c972ea7deb9/protobuf-3.6.1-cp36-cp36m-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl
Collecting grpcio>=1.8.6 (from tensorflow==1.10.0)
  Using cached https://files.pythonhosted.org/packages/50/11/e1a1e4912131c783f01734597150a58d75003d89aeff3ebf3d2ee5b4d36a/grpcio-1.18.0-cp36-cp36m-macosx_10_9_x86_64.whl
Collecting gast>=0.2.0 (from tensorflow==1.10.0)
Collecting werkzeug>=0.11.10 (from tensorboard<1.11.0,>=1.10.0->tensorflow==1.10.0)
  Using cached https://files.pythonhosted.org/packages/20/c4/12e3e56473e52375aa29c4764e70d1b8f3efa6682bef8d0aae04fe335243/Werkzeug-0.14.1-py2.py3-none-any.whl
Collecting markdown>=2.6.8 (from tensorboard<1.11.0,>=1.10.0->tensorflow==1.10.0)
  Using cached https://files.pythonhosted.org/packages/7a/6b/5600647404ba15545ec37d2f7f58844d690baf2f81f3a60b862e48f29287/Markdown-3.0.1-py2.py3-none-any.whl
Installing collected packages: setuptools, termcolor, six, absl-py, werkzeug, markdown, wheel, protobuf, numpy, tensorboard, astor, grpcio, gast, tensorflow
Successfully installed absl-py-0.7.0 astor-0.7.1 gast-0.2.2 grpcio-1.18.0 markdown-3.0.1 numpy-1.14.5 protobuf-3.6.1 setuptools-40.6.3 six-1.12.0 tensorboard-1.10.0 tensorflow-1.10.0 termcolor-1.1.0 werkzeug-0.14.1 wheel-0.32.3
Solving environment: ...working... done

## Package Plan ##

  environment location: /anaconda3/envs/r-tensorflow

  added / updated specs: 
    - h5py
    - pillow
    - pyyaml
    - requests
    - scipy


The following NEW packages will be INSTALLED:

    asn1crypto:      0.24.0-py36_1003                       conda-forge
    blas:            1.1-openblas                           conda-forge
    cffi:            1.11.5-py36h342bebf_1001               conda-forge
    chardet:         3.0.4-py36_1003                        conda-forge
    cryptography:    2.3.1-py36hdbc3d79_1000                conda-forge
    freetype:        2.9.1-h597ad8a_1005                    conda-forge
    h5py:            2.9.0-nompi_py36h35dc34b_1102          conda-forge
    hdf5:            1.10.4-nompi_h646315f_1105             conda-forge
    idna:            2.8-py36_1000                          conda-forge
    jpeg:            9c-h1de35cc_1001                       conda-forge
    libgfortran:     3.0.1-h93005f0_2                                  
    libpng:          1.6.36-ha441bb4_1000                   conda-forge
    libtiff:         4.0.10-h79f4b77_1001                   conda-forge
    numpy:           1.16.0-py36_blas_openblash4251c03_1000 conda-forge [blas_openblas]
    olefile:         0.46-py_0                              conda-forge
    openblas:        0.3.3-hdc02c5d_1001                    conda-forge
    pillow:          5.4.1-py36hbddbef0_1000                conda-forge
    pycparser:       2.19-py_0                              conda-forge
    pyopenssl:       18.0.0-py36_1000                       conda-forge
    pysocks:         1.6.8-py36_1002                        conda-forge
    pyyaml:          3.13-py36h1de35cc_1001                 conda-forge
    requests:        2.21.0-py36_1000                       conda-forge
    scipy:           1.2.0-py36_blas_openblash486cb9f_1201  conda-forge [blas_openblas]
    six:             1.12.0-py36_1000                       conda-forge
    urllib3:         1.24.1-py36_1000                       conda-forge
    yaml:            0.1.7-h1de35cc_1001                    conda-forge

The following packages will be UPDATED:

    certifi:         2018.11.29-py36_0                                  --> 2018.11.29-py36_1000  conda-forge
    tk:              8.6.8-ha441bb4_0                                   --> 8.6.9-ha441bb4_1000   conda-forge

The following packages will be DOWNGRADED:

    ca-certificates: 2018.12.5-0                                        --> 2018.11.29-ha4d7672_0 conda-forge
    openssl:         1.1.1a-h1de35cc_0                                  --> 1.0.2p-h1de35cc_1002  conda-forge
    python:          3.6.8-haf84260_0                                   --> 3.6.7-h4a56312_1001   conda-forge

Preparing transaction: ...working... done
Verifying transaction: ...working... done
Executing transaction: ...working... done


==> WARNING: A newer version of conda exists. <==
  current version: 4.4.10
  latest version: 4.6.1

Please update conda by running

    $ conda update -n base conda


Collecting keras
  Using cached https://files.pythonhosted.org/packages/5e/10/aa32dad071ce52b5502266b5c659451cfd6ffcbf14e6c8c4f16c0ff5aaab/Keras-2.2.4-py2.py3-none-any.whl
Collecting tensorflow-hub
  Using cached https://files.pythonhosted.org/packages/5a/76/18c9aaf24f1b888bafa40351493815050e100098b4782d634b307fb1ebc8/tensorflow_hub-0.2.0-py2.py3-none-any.whl
Collecting keras-preprocessing>=1.0.5 (from keras)
  Using cached https://files.pythonhosted.org/packages/fc/94/74e0fa783d3fc07e41715973435dd051ca89c550881b3454233c39c73e69/Keras_Preprocessing-1.0.5-py2.py3-none-any.whl
Requirement already satisfied, skipping upgrade: h5py in /anaconda3/envs/r-tensorflow/lib/python3.6/site-packages (from keras) (2.9.0)
Requirement already satisfied, skipping upgrade: pyyaml in /anaconda3/envs/r-tensorflow/lib/python3.6/site-packages (from keras) (3.13)
Requirement already satisfied, skipping upgrade: numpy>=1.9.1 in /anaconda3/envs/r-tensorflow/lib/python3.6/site-packages (from keras) (1.16.0)
Requirement already satisfied, skipping upgrade: scipy>=0.14 in /anaconda3/envs/r-tensorflow/lib/python3.6/site-packages (from keras) (1.2.0)
Requirement already satisfied, skipping upgrade: six>=1.9.0 in /anaconda3/envs/r-tensorflow/lib/python3.6/site-packages (from keras) (1.12.0)
Collecting keras-applications>=1.0.6 (from keras)
  Using cached https://files.pythonhosted.org/packages/3f/c4/2ff40221029f7098d58f8d7fb99b97e8100f3293f9856f0fb5834bef100b/Keras_Applications-1.0.6-py2.py3-none-any.whl
Requirement already satisfied, skipping upgrade: protobuf>=3.4.0 in /anaconda3/envs/r-tensorflow/lib/python3.6/site-packages (from tensorflow-hub) (3.6.1)
Requirement already satisfied, skipping upgrade: setuptools in /anaconda3/envs/r-tensorflow/lib/python3.6/site-packages (from protobuf>=3.4.0->tensorflow-hub) (40.6.3)
Installing collected packages: keras-preprocessing, keras-applications, keras, tensorflow-hub
Successfully installed keras-2.2.4 keras-applications-1.0.6 keras-preprocessing-1.0.5 tensorflow-hub-0.2.0

Installation complete.

and then when I detected python using the following code
```{r}
reticulate::py_discover_config("keras")
python: /usr/bin/python
libpython: /System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/config/libpython2.7.dylib
pythonhome: /System/Library/Frameworks/Python.framework/Versions/2.7:/System/Library/Frameworks/Python.framework/Versions/2.7
version: 2.7.10 (default, Jul 15 2017, 17:16:57) [GCC 4.2.1 Compatible Apple LLVM 9.0.0 (clang-900.0.31)]
numpy: /System/Library/Frameworks/Python.framework/Versions/2.7/Extras/lib/python/numpy
numpy_version: 1.8.0
keras: [NOT FOUND]

python versions found:
/usr/bin/python
/anaconda3/envs/r-tensorflow/bin/python

reticulate::py_discover_config("tensorflow")
python: /usr/bin/python
libpython: /System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/config/libpython2.7.dylib
pythonhome: /System/Library/Frameworks/Python.framework/Versions/2.7:/System/Library/Frameworks/Python.framework/Versions/2.7
version: 2.7.10 (default, Jul 15 2017, 17:16:57) [GCC 4.2.1 Compatible Apple LLVM 9.0.0 (clang-900.0.31)]
numpy: /System/Library/Frameworks/Python.framework/Versions/2.7/Extras/lib/python/numpy
numpy_version: 1.8.0
tensorflow: [NOT FOUND]

python versions found:
/anaconda3/envs/r-tensorflow/bin/python
/usr/bin/python
```
I have no clue even after I review what you discussed above, wonder if anyone could give me some suggestions. Thank you all.

sjwscjj picture sjwscjj on 28 Jan 2019

That was difficult to solve for me. The way I debugged was this: open a terminal and activate the conda environment conda activate r-tensorflow.

Make sure you are actually running the python version from within the environment (e.g. it shouldn't be another python installation on your system, check that the python version matches and maybe use where python to check where python is; mine was /home/nate/anaconda3/envs/r-tensorflow/bin/python).

Run python, try to import keras and/or tensorflow. I got this error: ImportError: libcublas.so.9.0: cannot open shared object file: No such file or directory. Classic cuda headache material. The issue was I had CUDA 9.2 installed, and it was looking for 9.0. Instead of set up symlinks, I just uninstall cuda 9.2 and reinstalled 9.2. Then installed cudnn.

I also found I had a ~/.bash_profile, which I deleted. According to the install intructions, this can cause problems. Lastly, I added the environment variables to ~/.profile as in the instructions. After rebooting, it finally worked! Took hours to figure out.

nateGeorge picture nateGeorge on 14 Mar 2019

Can this issue and #730 be related?

marioem picture marioem on 7 Apr 2019

Why this errors comes IN R console
model=keras_model_sequential()
Error: Python module tensorflow.keras was not found.

Detected Python configuration:

python: C:/Users/Rimal/AppData/Local/r-miniconda/envs/r-reticulate/python.exe
libpython: C:/Users/Rimal/AppData/Local/r-miniconda/envs/r-reticulate/python36.dll
pythonhome: C:/Users/Rimal/AppData/Local/r-miniconda/envs/r-reticulate
version: 3.6.12 (default, Dec 9 2020, 00:11:44) [MSC v.1916 64 bit (AMD64)]
Architecture: 64bit
numpy: C:/Users/Rimal/AppData/Local/r-miniconda/envs/r-reticulate/Lib/site-packages/numpy
numpy_version: 1.19.4

yagyarimal picture yagyarimal on 10 Dec 2020
Was this page helpful?
0 / 5 - 0 ratings

Related issues

masterlord99 picture masterlord99  Â·  6Comments
gsimchoni picture gsimchoni  Â·  3Comments
LarsHill picture LarsHill  Â·  6Comments
qade544 picture qade544  Â·  5Comments
fderyckel picture fderyckel  Â·  4Comments