So I recently got a new system and I need to go through all the hoops to get GPU support to work for Keras in R.
I followed the steps and it seemed everything worked until I ran it and then this happened:
> library(keras) > imdb <- dataset_imdb(num_words = 10000) Error: Installation of TensorFlow not found. Python environments searched for 'tensorflow' package: C:\Users\xxx\Anaconda3\envs\r-tensorflow\python.exe C:\Users\xxx\Anaconda3\python.exe C:\Users\xxx\Anaconda3\python.exe You can install TensorFlow using the install_tensorflow() function.
Currently it is a nightmare - just given the size of downloads and not really winning. I have followed this link: https://www.tensorflow.org/install/gpu and this one: https://medium.com/@abasauri92/a-hopefully-useful-guide-to-installing-gpu-version-of-keras-in-r-c6cdbf6928ba to try and get it to work (along with several other posts).
State of what I have:
*I am running 64bit Windows 10
*I have a NVIDIA GPU - RTX 2080 TI so my card is definitely compatible.
*I have R version 3.6.0 (2019-04-26) -- "Planting of a Tree"
*Anaconda - I have Anaconda3-5.3.1-Windows-x86_64 as the other versions were giving problem and this seemed to be one that would solve the issue (it did).
*For Cuda I downloaded and installed: cuda_10.1.168_425.25_win10
*For cuDNN I installed: cudnn-10.1-windows10-x64-v220.127.116.11 (and yes I did extract it to the correct folders).
*My Keras is version: 18.104.22.168.9001
After the above I ran this in the terminal:
install.packages(“devtools”) library(devtools) install.packages(“reticulate”) library(reticulate) install_github(“rstudio/keras”) library(keras) install_keras(tensorflow = “gpu”) reticulate::py_config() reticulate::py_module_available(“keras”)
All of it installed and worked fine until we got to the last two lines. It did return false and didn't work.
Any help would be appreciated on this topic!