Teaching computer to recognize railroads on maps

Instructions for installing deep learning environment on Microsoft Azure

1)

Install nvidia drivers (if necessary). Might have to check the most recent version Check for latest drivers !!! MAKE SURE YOUR KERNEL VERSION IS GREATER THAN 4.4.0-75 (IT HAS TO BE AT LEAST 4.4.0-77) !!!

(~12 minutes on NC6 [2017-04-25])

DRIVER_VERSION="375.66"
wget http://us.download.nvidia.com/tesla/${DRIVER_VERSION}/nvidia-diag-driver-local-repo-ubuntu1604_${DRIVER_VERSION}-1_amd64.deb

sudo dpkg -i nvidia-diag-driver-local-repo-ubuntu1604_${DRIVER_VERSION}-1_amd64.deb
sudo apt-get update
sudo apt-get install -y cuda-drivers

rm nvidia-diag-driver-local-repo-ubuntu1604_${DRIVER_VERSION}-1_amd64.deb

# Optional - restart your machine
# Optional - check the status of GPUs: nvidia-smi

2) Install Docker

https://docs.docker.com/engine/installation/linux/ubuntu/#install-using-the-repository (~1 minute on NC6 [2017-04-25])

# Install packages to allow `apt` to use a repository over HTTPS:
sudo apt-get install -y apt-transport-https ca-certificates curl software-properties-common

# Add Docker's official GPG key:
curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo apt-key add -


# Verify that the key fingerprint is `9DC8 5822 9FC7 DD38 854A E2D8 8D81 803C 0EBF CD88`
sudo apt-key fingerprint 0EBFCD88


# Set up apt repo:

sudo add-apt-repository \
   "deb [arch=amd64] https://download.docker.com/linux/ubuntu \
   $(lsb_release -cs) \
   stable"
   
sudo apt-get update

# Install latest version of Docker:
sudo apt-get install -y docker-ce

3) Install nvidia-docker and nvidia-docker-plugin

This automatically sets up all paths/libraries/environment variables necessary for a container started by sudo nvidia-docker run ... to access host's gpu(s)

LATEST_VERSION="1.0.1"
wget -P /tmp https://github.com/NVIDIA/nvidia-docker/releases/download/v${LATEST_VERSION}/nvidia-docker_${LATEST_VERSION}-1_amd64.deb
sudo dpkg -i /tmp/nvidia-docker*.deb && rm /tmp/nvidia-docker*.deb

4) Pull docker image from docker-hub

sudo nvidia-docker pull spatialcomputing/deep-learning-env-gpu

5) Run docker environment

sudo nvidia-docker run --rm -ti -p 8888:8888 -v /datadrive:/datadrive spatialcomputing/deep-learning-env-gpu

If you are not planning on running jupyter notebook from within the container, you don't need to pass -p host_port:container_port. If you are not planning on accessing host's file system, you don't need to pass -v host_directory_name:how_that_directory_will_appear_within_container_name, but remember that all files/data generated within running container will be lost when you stop it, unless it was written to host's directory.