[Q&A] Python Programming Pattern - Flare Client API vs Learner

[k2sebeom]

Python version (python3 -V)

3.9.6

NVFlare version (python3 -m pip list | grep "nvflare")

2.5.0

NVFlare branch (if running examples, please use the branch that corresponds to the NVFlare version, git branch)

No response

Operating system

Ubuntu 22.04

Have you successfully run any of the following examples?

• hello-numpy-sag with simulator
• hello-pt with simulator
• hello-numpy-sag with POC
• hello-pt with POC

Please describe your question

Hi! Just a simple question about programming patterns.

I have successfully run hello-pt and been fairly satisfied with flare.send(FLModel) pattern to send the result from client. Then, I moved on to MLFlow experiment tracking and found the client code has completely changed to inherit "Learner" and now each methods are returning DXO.

I am curious which pattern is more recommended when it comes to programming patterns. For example, is it possible to so cross-site validation using flare client api as well?

[YuanTingHsieh]

@k2sebeom thanks for the questions!

We have a lot of different levels of APIs that are targeting flexibility VS easy usability.

For people to start with, we recommend the Flare Client API way.
You CAN do cross-site validation using flare client api as well.
Please refer to: https://github.com/NVIDIA/NVFlare/tree/main/examples/hello-world/step-by-step/cifar10/cse
This example of "step-by-step" shows a complete picture of how to do the same cifar10 with different algorithms/APIs that NVFlare provides.

We are continuing organizing our examples/tutorials so it is easier for people to find the right materials.

Thanks!

[chesterxgchen]

Yes, we need to update the MLFLow example to use Client API instead of using Learner pattern.

Here is the example in the tutorials

https://github.com/NVIDIA/NVFlare/blob/main/examples/tutorials/self-paced-training/part-1_federated_learning_introduction/chapter-1_running_federated_learning_applications/01.5_experiment_tracking/01.5.2_experiment_tracking_with_mlflow/code/src/client_mlflow.py