Docker Research Images
This repository contains a collection of Docker images that I have used to set up my research environments. Inspired by the Deepo series of Docker images.
ML Research
Machine learning research environment including common tools for model development and data analysis. Built upon the TensorFlow Base images (PyTorch Base tags to be added 🙂). See image tags on Docker Hub for the TensorFlow versions that I am currently building/using. See also the included libraries and their versions in ml_research/<image_tag>/requirements.txt.
Run Docker Environment Script
I have included a script run_docker_env.sh that I use to spin up docker containers. This is useful with my reloff/ml-research image to quickly setup machine learning research environments without the hassle of messing with Docker parameters to mount folders, set file ownership, etc. This inlcudes optional sudo with --sudo, so that your files are owned by you by default, access to the nvidia-docker GPU runtime with a simple --nvidia-gpu flag, access to Jupyter Lab or Notebook, easy volume mounting, and other convenience options.
To install on a local level, run ./install_docker_env.sh. You should then have access to the docker environment script run-docker-env from anywhere on your machine.
See run-docker-env --help for information on arguments.
TensorFlow Base
Provides a minimal installation of TensorFlow (https://www.tensorflow.org/) (1.x and 2.x) with GPU support on Ubuntu, intended to be used as a base image for TensorFlow research environments.
Currently provides only Ubuntu 16.04 (with CUDA+cuDNN) and Python 3.6 Docker images. Requires nvidia-docker (https://github.com/NVIDIA/nvidia-docker) to run containers with GPU support.
PyTorch Base
Provides a minimal installation of PyTorch (https://pytorch.org/), intended to be used as a base image for PyTorch research environments.
Currently provides only Ubuntu 16.04 (with CUDA+cuDNN) and Python 3.6 Docker images. Requires nvidia-docker (https://github.com/NVIDIA/nvidia-docker) to run containers with GPU support.
Kaldi
Provides the Kaldi Speech Recognition Toolkit (http://kaldi-asr.org) in a simple Docker image.
Kaldi is currently built from fork https://github.com/rpeloff/kaldi/tree/5.4 on the official Ubuntu 16.04 Docker image. Source and tools are located at /kaldi.