Docker image for the creation of portable Linux Conda environment built using conda-pack.

This tool requires a local installation of Docker and Python 3.

First, download or clone the repository:

git clone [email protected]:mcencini/conda-builder.git

Navigate to the repository main folder and enter the src folder to build the docker image:

cd /path/to/repo/src
docker build -t conda-builder:alpha .

We provide a create_packed_environment.py Python script to run the docker command. Usage is:

python /path/to/folder/create_packed_environment.py args

Here, args represents the following positional arguments:

path-to-config.yaml  - this is the path to conda environment YAML config file.
environment-name (optional) - this is the resulting environment name. Defaults to "env".
output-path (optional) - this is the path where the packed environment is saved. Defaults to the parent folder of "config.yaml"

For example, creation of a Pytorch environment using the provided configuration example can be done as:

python /abs-path-to/create_packed_environment.py /abs-path-to/examples/torch_environment.yaml "pytorch-cpu"

This will create a pytorch-cpu folder containing a pytorch-cpu.tar.gz archive and a unpack_pytorch-cpu.sh shell script. In general, the script will create an environment-namefolder containing a environment-name.tar.gz archive and a unpack_environment-name.sh script.

To relocate the environment in the target machine, copy the newly created folder in the desired location and run the following commands:

cd /abs-path-on-target-machine/environment-name
./unpack_environment-name.sh

This will untar the environment, clean the prefixes and create a shortcut to the activation script in the environment parent folder. Specifically, the path on target machine will appear as:

/abs-path-on-target-machine/
├── environment-name/ 
└── activate-environment-name

That's it! Now, the environment can be activated when desired by running:

source /abs-path-on-target-machine/activate-environment-name

Some caveats:

• Currently, we do not provide a shortcut to the deactivation script. Environment can be deactivated by running source /abs-path-on-target-machine/environment-name/bin/deactivate.
• Activation does not work with csh shell due to limitations in Conda activate script.