9dogs / ml-env

Environment for machine learning experiments.

ml-env

Conda-based environment for machine learning experiments.

Image has 2 flavors: CPU (~1.7 Gb) and GPU (~2.6 Gb). Pull 9dogs/ml-env:latest for GPU-enabled version and 9dogs/ml-env:cpu for CPU-only version.
For GPU image to work properly NVIDIA driver and Docker > 19.03 must be installed (and you have to have Linux host).

Feel free to adjust environment.yml and additional_packages.txt to fit your needs.

Run

To run JupyterLab:

• Windows: docker run --rm -it -p 4545:4545 -v %cd%:/notebooks -w /notebooks 9dogs/ml-env lab
• Linux: docker run --rm -it -p 4545:4545 -v $PWD:/notebooks -w /notebooks 9dogs/ml-env lab
• Linux with GPU: docker run --rm -it -p 4545:4545 --gpus all -v $PWD:/notebooks -w /notebooks 9dogs/ml-env lab

JupyterLab and Jupyter Notebook can be accessed on http://localhost:4545.

Or use run_docker_jupyter.py helper:

usage: run_docker_jupyter.py [-h] [--docker_tag DOCKER_TAG] [--gpus GPUS]
                             command

Run docker image.

positional arguments:
  command               Command (notebook | lab | shell)

optional arguments:
  -h, --help            show this help message and exit
  --docker_tag DOCKER_TAG, -t DOCKER_TAG
                        Docker image tag
  --gpus GPUS           GPUs to forward to the container (all | 1 | 2 etc.)

Build

To build an image, run: docker build -t <tag> . (where <tag> can be anything, e.g. ml:latest).
To add packages modify additional_packages.txt or environment.yml (slower) files then rebuild the image.