Dockerfile for Machine Learning

A Dockerfile for accelerated research process, consists of major machine learning libraries.

• DockerHub

Features

• Ubuntu 18.04
• Python 3.6 (Miniconda 3)
• GPU accelerated (CUDA 10.0, cuDNN 7)
• NCCL, CNMeM, Apex (PyTorch only) activated
• Jupyter and OpenCV 3.0 included
• Additional packages (Tensorboard, Hyperdash, etc)

Available Deep Learning Frameworks

• TensorFlow uetchy/ml:tensorflow
• PyTorch uetchy/ml:pytorch
• Chainer uetchy/ml:chainer
• MXnet uetchy/ml:mxnet
• XGBoost uetchy/ml:xgboost

Install

System Requirements

• Docker
• CUDA-enabled GPUs
• CUDA Driver
• CUDA Toolkit
• nvidia-docker2 (https://github.com/NVIDIA/nvidia-docker)

Pull the docker image from DockerHub

docker pull uetchy/ml:tensorflow
docker pull uetchy/ml:pytorch
docker pull uetchy/ml:chainer
...

Launch Jupyter Lab on current directory

docker run --runtime=nvidia -v $PWD:/workspace -p 8888:8888 -it uetchy/ml:pytorch jupyter
open http://localhost:8888

Open Python REPL

docker run --runtime=nvidia --rm -it uetchy/ml:base python

Run Bash Shell

docker run --runtime=nvidia --rm -it uetchy/ml:tensorflow

List of Docker images for Data Science

• TensorFlow
• Caffe2
• MXNet
• Caffe

Contribution

PRs are accepted.

Contributors

• Yasuaki Uechi
• UpmostScarab
• cyrusmvahid