ml-docker

Just another custom docker environment for machine learning and data science projects. Cuda in docker is also supported.

docker - machine learning - environment - cuda

Author: Maximilian Bundscherer

Overview

scripts/: Scripts included
docker/: Docker stuff included

You can use jupyter lab or run python scripts in attached custom docker container.

Let's get started

Navigate to scripts folder cd scripts/
Build docker container ./build-docker.sh

Jupyter lab mode

Do all steps in Let's get started (above)
Start docker container ./start-jupyter-lab.sh ~/Desktop
- Replace ~/Desktop by repo-path without / at end
Go to http://127.0.0.1:50888/lab
- Login with token abc

Script mode

Do all steps in Let's get started (above)
Start docker container ./start-script.sh ~/Desktop script.py
- Replace ~/Desktop by repo-path without / at end
- Replace script.py by script filename in repo-sub-path
Read log cat run.log

A word about docker and cuda

If you want to use cuda inside of a docker container, start nvidia-docker instead of docker
If you want to specify which gpus you want to use, add prefix NV_GPU=3,4 (check gpus on server with nvidia-smi before)
The scripts in the repo has built-in cuda support (scripts/start-jupyter-lab.sh and scripts/start-script.sh) (but please change which gpus you use)

Debug section

Navigate to scripts folder cd scripts/
./check-docker.sh: Check if docker container is running and check if image is available
./remove-docker.sh: Remove running docker container and remove image

Check if torch can use gpus

import sys

import torch

print(f"PyTorch Version: {torch.__version__}")
print()
print(f"Python {sys.version}")
print("GPU is", "available" if torch.cuda.is_available() else "NOT AVAILABLE")

maxbundscherer/ml-docker