Feedback/info request on Tensor Flow deeplearning



Hey ! Is there anyone out who is working with Tensorflow. Like to get some feedback/info on usage of tensorflow especially for deep learning. Is working with Tensorflow has some advantage over R.
I understand that implementing algo trained & tested on R is difficult task, infact impossible for algos like xgboost. What is situation with TensorFlow.
Should I start learning python for tensorFlow.
Does TensorFlow has some advantage like ease to implement or some other over R.
Please show me some light if anyone experienced both.


Hello @ParindDhillon,

A small note; With latest frameworks like H2O and MXNetR, the roadblocks for building deep learning models in R practically vanish. You should really check them out.

Regarding Tensorflow, I am currently trying to go through it (in python) and I say it is an excellent library. It has support for python and C++ at the moment, but they propose to develop interfaces for other languages in the near future (including R). Some of the advantages (as per their official site) are:

  • Flexibility: Tensorflow represents computation on a data flow graph. This graph can be changed according to your needs.
  • Portability : It supports many platforms along with support for CPUs/GPUs. You can deploy the models on mobile too!
  • Distributed nature : It handles the backend on its own (like threads, queues, etc), abstracting the user from its complexity, so that you can efficiently use the hardware.

In short, you could use tensorflow when you have many different resources at hand, and you want to handle them distributively and as efficiently as possible.

What I would recommend you is, if you are more comfortable in R, try using vanilla MXnet or H2O, or integrate H2O with tensorflow.


Thanks JalFaizy thats pretty detailed reply. I have used H20 couple of times & heard about MXNet few days back. h20 is great & mxnet looks promising, but i am bit under-confident in implementing real time algos using R. I have worked on various datasets for contest purposes, but in case of production environment, dont have much experience. Just because I could not implement xgboost in production with 85% variance explanation, I had to use regression with 75% variance explanation, regression gave me parameters & easier for my start up to implement. I feel R lacks production environment usage. Please correct me if i am missing anything. I think python might be easier to use & hence tensorFlow too. Anyways thanks once again.


Personally I have not done a lot of R programming. And I have not (yet) deployed a machine learning model, so I may not be the right person to talk about “production environment”.

Still I must say that in production, reliability is important. Its the same reason Java is preferred over python right? (Also, you have someone to blame if something goes wrong :wink: ) But it should not be the reason to shift over languages and starting from the scratch unless you are absolutely sure and have done extensive survey of current techniques.

PS: This presentation does a SWOT analysis on deep learning frameworks for production.