Skripsi

Prerequisites

• NVIDIA GPU (Tesla T4 or equivalent)
• Ubuntu 18.04
• CUDA Toolkit 10.0
• Docker
• nvidia-docker

Setup and start

Create and move to temporary folder:

mkdir /temp
cd /temp

Clone dockerfiles repository:

git clone https://github.com/eufat/dockerfiles.git
cd dockerfiles/jupyter-keras-gpu

Build docker image as jupyter-gpu tag:

nvidia-docker build -t jupyter-gpu .

Clone this repository to root directory:

cd /root
git clone https://github.com/eufat/skripsi.git

Run jupyter-gpu image as skripsi container and mount it to host directory:

nvidia-docker run -it -d \
    --mount type=bind,source=/root/skripsi/,target=/notebooks/skripsi \
    -p 8888:8888 -p 6006:6006 \
    --name skripsi \
    jupyter-gpu

Start the skripsi container you have built:

nvidia-docker start skripsi

(Optional: add nvidia-docker start skripsi command to your remote server startup script)

JupyterLab

Open localhost:8888/lab or server_ip_address:8888/lab when using remote server (allow port 8888 in remote server firewall first).

TensorBoard

Run tensorboard --logdir=/notebooks/logs --host 0.0.0.0 inside container or in JupyterLab terminal. Open localhost:6006 or server_ip_address:6006 when using remote server (allow port 6006 in remote server firewall first).

Pulling datasets

Install Git LFS first:

curl -s https://packagecloud.io/install/repositories/github/git-lfs/script.deb.sh | sudo bash
sudo apt-get install git-lfs

Inside skripsi repository, pull .raw HSI datasets:

git lfs install
git lfs pull