An example of building a TensorFlow model from R using rPython

tensorflow_in_r_using_rpython.R

### An example of building a TensorFlow model from R using rPython ###
# For this script you need to 
# 1. Have python 2.7 installed.
# 2. Install the rPython package in R.
# 3. Install Google's TensorFlow library as per these instructions:
#     http://www.tensorflow.org/get_started/os_setup.md#binary_installation

### Here is how to setup and run a trivial TensorFlow model ###

# Load TensorFlow (I couldn't get this to work without setting sys.argv... )
library(rPython)
python.exec("
import sys
sys.argv = ['']
import tensorflow as tf
")

# Define a "hello world" TensorFlow model adding two numbers
python.exec("
a = tf.constant(10)
b = tf.constant(32)
sum = a + b
")

#Instantiate a TensorFlow session, and get the result into R.
# (we need the .tolist() to convert from the result into something
# that can be serialized by JSON and imported into R)
python.exec("
sess = tf.Session()
result = sess.run(sum)
")
result = python.get("result.tolist()")

# Tada! :)
print(result)
## [1] 42

This is nice example. I have been trying to do something similar but am running into a "cannot import name pywrap_tensorflow" error when trying to import tensorflow. I know this is typically a circular reference error, but I haven't been able to figure out what is causing it. I was just curious if you ran into it when getting things set up.

Thanks!

i get a problem when i have huger data sets. it breaks execution
Fehler in python.exec("\n for i in range(10):\n sess.run(train_step, feed_dict={x: trainX, y_: trainY})\n ") :
����

No module named tensorflow.

I get this error when I run tensorflow from R. However from the python console it is running fine. What can be the reason?