Docs

Prepare Learner Workers

Mandatory

FLOps’ main goal is to perform federated learning, i.e., to train a machine learning model on local data in a distributed and privacy-preserving way. Training an ML model requires compatible data. A worker node is required to aggregate matching data to be able to participate in training. By default, orchestrated nodes do not aggregate training data. Only worker nodes that have been prepared as described on this page can collect such data and become FL learners.

For FLOps to work as intended, you are required to prepare at least one of your orchestrated nodes, as described in this guide.

Optimize Learning

Training is the centerpiece of FLOps, not only conceptually but also computationally and runtime-wise. Select your learner nodes wisely. Prefer more powerful, resource-rich machines to speed up training times.

How do learners handle and store data for training in FLOps?

Explore how FLOps manages ML data for local training here

On the worker nodes where you wish to perform ML model training, do the following:

  • Ensure the NodeEngine is running
      sudo NodeEngine -a <cluster-address> -d && sudo NodeEngine status
  • Activate the FLOps-learner addon for the NodeEngine:
      sudo NodeEngine config addon FLOps-learner on
  • Restart the NodeEngine
    • Either run sudo NodeEngine stop and then start it up again
    • Or run sudo systemctl restart nodeengine.service
  • Verify that the addon is active:
      > sudo NodeEngine config addon

  Configured Addons:
       - FLOps-learner: 🟢 Active
Prev
Prepare Image-builder Workers
Next
FLOps Projects Overview