Algorithms

---

This repository is part of vantage6, our privacy preserving federated learning infrastructure for secure insight exchange. The aim of this repository is to collect algorithms, ensure quality and continues delivery in our Docker image registry (harbor2.vantage6.ai).

Algorithm documentation: algorithms.vantage6.ai

Algorithms

Generalized Linear Model (GLM)

image: harbor2.vantage6.ai/algorithms/glm:latest
test-image: harbor2.vantage6.ai/algorithms/glm:[SHORT_COMMIT_HASH]
repository location: /models/glm
langauge: R

---

...

image: ...
test-image: ...
repository localtion: /
langauge: ...

---

Algorithm Submission Guidelines

Some general notes:

• Max characters per line: 80
• Use when using Python and use when using R
• Every algorithm should have its own workflow file in .github/workflows to build it
• Non-production algorithms should be tagged with the short commit hash
• Document the algorithm in the /docs folder using ReST

_ We are developing a cookiecutter which helps you to kickstart your algorithm project_

R-algorithm

Style: Tidyverse

Template: ... (TODO)

In R the package is installable (providing additional client side tooling and because we need to install it into the Docker image), therefore the package should have at least the following structure:

glm
│   .gitignore
│   project.Rproj
│   README.md
│   LICENSE
│
├───docker
│       Dockerfile
│
└───src
    │       DESCRIPTION
    │       DESCRIPTION.tpl
    │       Makefile
    │       NAMESPACE
    │
    ├───data
    │       ...
    │
    ├───man
    │       ...
    │
    └───R
            dglm.mock.R
            dglm.R

The man folder contains generated documentation. Make sure you document all your functions using Roxygen style. data should contain some data that can be used to test the method using dglm.mock.R. The DESCRIPTION is generated from DESCRIPTION.tpl using the build instruction in the makefile. The NAMESPACE file should be there, just copy it from the template.

Notes

• Use devtools to create you package: github
• You can add data by using the method devtools::use_data(...)
• If you add package data make sure you document it in R/data.R

Python Algorithm

Style: PEP 8

Template: https://github.com/IKNL/v6-boilerplate-py

In python the algorithm needs to be installable into the Docker image. The minimal folder structure:

pkg_name
│   .gitignore
│   Dockerfile
│   LICENSE
│   README.md
│   requirements.txt
│   setup.py
│
└───v6-boilerplate-py
    │   example.py
    │   __init__.py
    │
    └───local
            data.csv
            input
            output
            token