Local Coiled software environments

This article will show you how you can use an environment.yml file to duplicate one of the default Coiled software environments.

Coiled documentation contains additional information on using software environments with Coiled. Coiled also maintains a number of default software environments.

While it is often possible to use coiled install <account>/<sofware-environment> to create a coiled sofware environment locally that contains the same dependencies as the coiled software environment, it is also easy to do this with an environment.yml file and conda.

Using an environment.yml file

On the Coiled software environments page, when you click on the ‘eye’ icon, you can view a Python dict with the conda channels and dependencies for each environment.

Viewing Software Environment Dependencies

Coiled Software Envionments (click image to enlarge)

That can then be pasted into a script like this to create an environment.yml file:

coiled_env_name = "coiled-default"
coiled_env = {
    "channels": ["conda-forge", "defaults"],
    "dependencies": [

with open("environment.yml", "w") as output:
    output.write(f"name: {coiled_env_name} \n")
    output.write("channels: \n")
    for channel in coiled_env["channels"]:
        output.write(f"  - {channel} \n")
    output.write("dependencies: \n")
    for package in coiled_env["dependencies"]:
        output.write(f"  - {package} \n")

Which creates output like this:

name: coiled-default
  - conda-forge
  - defaults
  - bokeh>=2.1.1
  - bottleneck
  - cytoolz
  - dask-image>=0.3.0
  - dask-ml>=1.5.0
  - dask=2021.5.0
  - h5py
  - lz4
  - numba
  - numpy>=1.19.0
  - pandas>=1.1.0
  - pillow>=7.2.0
  - pip
  - pyarrow>=0.15.1
  - python-blosc
  - python-graphviz
  - python=3.9
  - requests
  - s3fs
  - scikit-learn>=0.23.1
  - xarray

Then, simply run conda env create -f environment.yml from a terminal (or Windows command prompt) to have conda create that sofware environment.

Note that you will still need to add any code-specific dependencies you might have, and because not all of the dependencies are pinned, it is still possible for client = Client(cluster) to report version mismatches. You should update your local packages accordingly.