A privacy preserving federated learning solution

---

v6-colnames-py

This repository contains a Vantage6 algorithm that collects the column names of the datasets made available by all organizations in a collaboration and returns:

• the column names per organization
• the sorted intersection of column names in common

This version targets Vantage6 4.14.0.

What changed for Vantage6 4.14.0

The algorithm was upgraded from the legacy 3.x container interface to the 4.x algorithm-tools interface:

• the Docker image now uses vantage6.algorithm.tools.wrap.wrap_algorithm()
• the algorithm uses @algorithm_client and @data() decorators
• child tasks are created with client.task.create(...)
• results are collected with client.wait_for_results(...)
• the local mock test uses MockAlgorithmClient

Dockerfile contract

For Vantage6 4.x, the image only needs Python plus the wrap_algorithm entrypoint. This repository therefore uses a self-contained Dockerfile and does not depend on harbor2.vantage6.ai/algorithms/algorithm-base:

FROM python:3.10-slim

ARG PKG_NAME="v6-colnames-py"

RUN apt-get update && \
    apt-get install --yes --no-install-recommends \
        build-essential \
        gfortran \
        libopenblas-dev \
        liblapack-dev \
        pkg-config && \
    rm -rf /var/lib/apt/lists/*

RUN pip install --upgrade pip setuptools wheel && \
    pip install \
        "openpyxl>=3.0.0" \
        "pandas>=1.5.3" \
        "pyfiglet==1.0.4" \
        "pyjwt==2.12.1" \
        "SPARQLWrapper>=2.0.0" \
        "sqlalchemy==1.4.46" \
        "vantage6-common==4.14.0" \
        "vantage6-algorithm-tools==4.14.0"

COPY . /app
RUN pip install --no-deps /app

ENV PKG_NAME=${PKG_NAME}

CMD python -c "from vantage6.algorithm.tools.wrap import wrap_algorithm; wrap_algorithm()"

Local test

The local test uses the 4.x mock client and in-memory pandas dataframes:

python -m pytest test.py

Example researcher script

The run.py file contains an example of creating a task from the Python client using the 4.14.0 API:

python run.py

You will need to adapt the server URL, credentials, collaboration, and image name to your own infrastructure.

Result shape

The master task returns a JSON-serializable object like:

{
  "organization_101": {
    "organization_id": 101,
    "node_id": 11,
    "columns": ["age", "patient_id"]
  },
  "organization_202": {
    "organization_id": 202,
    "node_id": 22,
    "columns": ["age", "patient_id", "site"]
  },
  "common": ["age", "patient_id"]
}

Read more

• Documentation: https://docs.vantage6.ai/
• Source code: https://github.com/vantage6/vantage6