This section will cover the installation of MXNet in R.

The MXNet package is a lightweight deep learning architecture supporting multiple programming languages such as R, Python, and Julia. From a programming perspective, it is a combination of symbolic and imperative programming with support for CPU and GPU.

The CPU-based MXNet in R can be installed using the prebuilt binary package or the source code where the libraries need to be built. In Windows/mac, prebuilt binary packages can be download and installed directly from the R console. MXNet requires the R version to be 3.2.0 and higher. The installation requires the drat package from CRAN. The drat package helps maintain R repositories and can be installed using the install.packages() command.

To install MXNet on Linux (13.10 or later), the following are some dependencies:

• Git (to get the code from GitHub)
• libatlas-base-dev (to perform linear algebraic operations)
• libopencv-dev (to perform computer vision operations)

To install MXNet with a GPU processor, the following are some dependencies:

• Microsoft Visual Studio 2013
• The NVIDIA CUDA Toolkit
• The MXNet package
• cuDNN (to provide a deep neural network library)

Another quick way to install mxnet with all the dependencies is to use the prebuilt Docker image from the chstone repository. The chstone/mxnet-gpu Docker image will be installed using the following tools:

• MXNet for R and Python
• Ubuntu 16.04
• CUDA (Optional for GPU)
• cuDNN (Optional for GPU)

1. The following R command installs MXNet using prebuilt binary packages, and is hassle-free. The drat package is then used to add the dlmc repository from git followed by the mxnet installation:

install.packages("drat", repos="https://cran.rstudio.com")
drat:::addRepo("dmlc")
install.packages("mxnet")

2. The following code helps install MXNet in Ubuntu (V16.04). The first two lines are used to install dependencies and the remaining lines are used to install MXNet, subject to the satisfaction of all the dependencies:

sudo apt-get update
sudo apt-get install -y build-essential git libatlas-base-dev
libopencv-dev
git clone https://github.com/dmlc/mxnet.git ~/mxnet --recursive
cd ~/mxnet
cp make/config.mk .
echo "USE_BLAS=openblas" >>config.mk
make -j$(nproc)

3. If MXNet is to be built for GPU, the following config needs to be updated before the make command:

echo "USE_CUDA=1" >>config.mk
echo "USE_CUDA_PATH=/usr/local/cuda" >>config.mk
echo "USE_CUDNN=1" >>config.mk

4. The following command is used to install MXNet (GPU-based) using Docker with all the dependencies:

docker pull chstone/mxnet-gpu