I'm asking this question just because currently I have learned R basics, dplyr, data.table, ggplot2, ggvis and machine learning packages such as caret, e1071 package for classification and nnet and neuralnet packages. The problem I faced with R is mainly related to deep learning.
Even though currently deep learning R comparable packages are available but they are not properly optimized for R. Such as H2o has great compatibility with R and has great feed forward deep learning support but it still lacks in convolution networks. Currently keras and mxnet has also provided deep learning supports but still it lacks many aspects of scikit-learn and tensorflow features. Though tensorflow support has been provided in the starting of 2017 for R still the syntaxes are quite fuzzy and pythonic with respect to R code structures.
In the context of deep learning, is there any scope available for R programming or still just to learn deep learning I have to learn python? Any suggestion would be appreciated.
Yes, You can definitely use R to learn/practice Deep Learning. For example, have a look at this webpage:
https://mxnet.incubator.apache.org/
or this one:
http://docs.h2o.ai/h2o/latest-stable/index.html
I think you can use R programming can solve many deep learning problems. I had some experiences on using R & Python to construct deep learning. Although many experiments on the internet almost using python, but R increases quickly and will become the most popular for whom willing to apply deep learning. In many deep learning problems, we only use R alone to solve easily.
Mxnet is a strong package for deep learning. Besides, you can consider darch, deepnet, H2O and deepr, ...
Thanks everyone for your valuable answers. From now onward, I will more concentrate on R and after getting a strong foothold on R programming, I would go for python.
Hello,
here's an answer from a R user. R is a great and is very capable and addictive. If you get a hold of deep learning in R many concepts will become clear to you. However major drawbacks in my opinion are:
1. R is slow (can be overcome with C/FORTRAN code which can be seamlessly integrated).
2. R is designed to be single threaded,(can be overcome with packages such as parallel and foreach).
3. R has very limited (read: next to none) GPU support (some indications exist that it may develop over time).
Once you learn the "whats and wheres" you will want to train your models fast so you can fiddle with them. R is not good for fast training.
That being said, no one language is king. Python has some nice libraries and some of them provide GPU support. Torch7 looks quite promising: http://torch.ch/. If I could wish I knew one more programming language it would be C.
You have to start somewhere.
All the best,
Milan
This is the interface to tensor flow from R
https://tensorflow.rstudio.com/
Yes, and there is even a book about doing just that:
https://www.manning.com/books/deep-learning-with-r
- Badges
- Science topic