Model packaging

Motivation...

Templating allows us to ...

Because we've chosen the approach of having a single docker image representing our body of models,

Requirements

• Must include lume-services

steps

1. Create package repository

.json template

"{{cookiecutter.repo_name}}"
├─ .github/
|  ├─ workflows/
|  |  ├─ build_flow_docker.yaml
│  │  ├─ tests.yaml
├─ "{{cookiecutter.project_slug}}"
|  ├─ files/
|  |  ├─ __init__.py
|  |  ├─ variables.yaml
|  ├─ flow/
|  |  ├─ __init__.py
|  |  ├─ _entrypoint.sh
|  |  ├─ Dockerfile
|  |  ├─ flow.py
|  ├─ tests/
|  ├─ __init__.py
|  ├─ config.py
|  ├─ model.py
├─ dev-requirements.txt
├─ requirements.txt
├─ README.md
├─ MANIFEST.in
├─ setup.cfg
├─ setup.py
├─ model.yaml

2. Define your model variables

Open your-project/your_project/files/variables.yml. The file will look like:

3. Define your model evaluation method

This class is extensible and can accomodate as many additional methods as you'd like to include.

4. Define your flow

A minimal flow will accept ...

Execution of a flow is defined in the blurb:

Where task1, task2, and task3 are defined in the module body using the @task decorator.

Flow targets: 1. Filesystem target 2. Mongodb target

Flows are also extensible and can accomodate plenty of complexity. Using database targets requires a configured resources to be available to the flow at runtime.

The README.md file should contain a comprehensive outline of variables accepted by your flow.

5. Actions

Define GitHub secret variables

Caveats

Because the repository is heavily templated, there are several things that may break on modification. The tests defined in the your-project/your_project/tests file are designed to test the following conditions: 1. Clearly defined entrypoints 2. Properly formatted and registerable flows 4.

JSON FILE WITH THE FOLLOWING:

{
  "author": "Jacqueline Garrahan",
  "email": "jacquelinegarrahan@gmail.com",
  "github_username": "jacquelinegarrahan",
  "project_name": "My Package", # Python importable package
  "project_slug": "{{ cookiecutter.project_name.lower().replace(' ', '_').replace('-', '_') }}",


  "project_short_description": "Python Boilerplate contains all the boilerplate you need to create a Python package.",
  "pypi_username": "{{ cookiecutter.github_username }}",
  "version": "0.1.0",
  "use_pytest": "n",
  "use_black": "n",
  "use_pypi_deployment_with_travis": "y",
  "add_pyup_badge": "n",
  "command_line_interface": ["Click", "Argparse", "No command-line interface"],
  "create_author_file": "y",
  "open_source_license": ["MIT license", "BSD license", "ISC license", "Apache Software License 2.0", "GNU General Public License v3", "Not open source"]
}

docker load --input /tmp/myimage.tar