Cookiecutter StepWorkflow

AICS Cookiecutter template for a simple data + code workflow:

• git(hub) for code
• quilt for data
• prefect to combine

An Example Workflow Produced with this Template

Getting started with this template

To use this template for a new workflow, use the following commands and then follow the prompts from the terminal.

pip install cookiecutter
cookiecutter gh:AllenCellModeling/cookiecutter-stepworkflow

Configuring your new project

Once you've followed the prompts, you should have a template repository that we need to

• install as a Python package
• connect to GitHub
• connect to Quilt

Install as a Python package

First, we'll make a conda environment to house this project's python dependencies. If you don't have conda installed, install it with miniconda.

Whatever you named your project, make a conda environment of the same name

conda create --name <project_name> python=3.7

and activate it with

conda activate <project_name>

To install the project as a python package, cd into the project directory, and then

cd <project_name>
pip install -e .[dev]

This will install your package in editable mode with all the required development dependencies.

Connect to GitHub

Create an empty repository on GitHub that has the same name as your project (you need to do this via the GitHub website). Don't initialize it with a README or anything.

Once the GitHub repo is created, push your project up to Github with

git remote add origin [email protected]:AllenCellModeling/<project_name>.git
git push -u origin master

If you get permissions errors, make sure you have ssh keys installed, or use https://github.com instead of [email protected]: in the origin address above.

Your initial commit will show a broken build badge. To fix this, configure codecov and a documentation generation access token following the instructions here.

Connect to Quilt

Access to quilt data in S3 requires two files

~/.aws/credentials:

[default]
aws_access_key_id=<your_secret_access_key_id>
aws_secret_access_key=<your_secret_access_key>

~/.aws/config:

[default]
region=us-west-2

Running your workflow

Example Step

This template comes with an example first workflow step Raw.

Run

You should be able to run this with the command

<project_name> raw run

This will write out some "raw data" (some randomly generated images) to local_staging/raw.

You should edit the run function of the Raw class in <project_name>/steps/raw/raw.py to do something relevant to your workflow, e.g. aggregating raw data and getting it ready to push to Quilt.

Push

To push the data in local_staging to quilt, use

<project_name> raw push

If your git branch is on master, this will save your data in quilt to aics/<project_name>/master/raw.

Checkout

To download the remote data and overwrite your local data, use

<project_name> raw checkout

Pull

To download the remote data needed as input to run a step, use

<project_name> raw pull

Since Raw is the first step, and doesn't need any inputs, this doesn't do anything here.

Add a new step

To make a new step in your workflow, in the main project directory use

make_new_step <StepName>

This will create a StepName class in <project_name/steps/step_name/step_name.py>, with a run method that is ready for you to edit.

If your step directly depends on the output of another step for input data to this step, set the direct_upstream_tasks kwarg in the class __init__ method to be a list of the steps this one depends on. The list should be of step classes, e.g. direct_upstream_tasks = [Raw].

For your step to run successfully, you need to save a dataframe manifest of the files you're writing out to self.manifest, and then save that as manifest.csv. See the Raw step for an example.

Run everything at once

To run all of your steps at once, use

<project_name> all run

push and checkout also work with all this way, to push or checkout all of your data at once.

If you add a new step to your workflow, you should also edit <project_name>/bin/all.py and in the All class, change self.step_list to include your new steps, in the order in which you want to run them.

Branches

You won't be able to push data to Quilt unless your git status is clean. This is intended to maintain parity between the data we save, and the code that generated it. To have alternate version of workflow data, just switch to a new git branch

git checkout -b <new_branch_name>

Pushing data to quilt with e.g. <project_name> push raw will then save your data to aics/<project_name>/<new_branch_name>/raw.

Optional configuration:

See the README here for all of the optional infrastructure you can (and should) add, e.g. docs, testing, etc.