`PipEnv`

Snakemake prefers Conda for Python environment management. Conda, however, is not appropriate for certain environments, such as compute clusters. PipEnv allows you to create traditional virtual environments on the fly, using virtualenv under the hood.

Start by creating a new environment:

from snakeboost import PipEnv

my_env = PipEnv(
  root="/tmp",
  packages=[
    "numpy",
    "pandas",
    "scipy==1.8.0",
    "git+https://github.com/nipy/nibabel@7cfaebf",
    "flake8"
  ]
)

Here, the root tells Snakeboost where to put the environment. packages is a list of valid package specifier. Version specifiers, git links, local files, anything valid in pip is valid here. You could also use the requirements=[...] parameter, passing a list of requirements.txt files to load.

Once created, use the venv in a rule:

rule lint_python:
    inputs: "some-script.py"
    shell:
        my_env.script("flake8 {input}")

This will run the flake8 command installed into your venv. Other methods include:

my_env.python(...): Execute a python file (e.g. script.py) or run a python module (e.g. "-m module") using the venv python.
my_env.make_venv(...): Create the venv, then execute any arbitrary command. This is helpful if you need access to your venv somewhere in the middle of the script. In this case, you could use the bin or python_path attributes within the script.

Note that the venv is never “activated”. So just having python <command> somewhere in your script won’t work. Use one of the PipEnv methods or attributes to get access to the exact path. This allows you to use as many PipEnvs as you want in one workflow, or even one rule!

Individual environments will be saved in a folder named with the hash of all package names and requirements.txt paths, and will be reused when possible. Environment creation is threadsafe, so having 32 rules using the same environment causes no issues.

See the full docs here

PipEnv

`PipEnv`