This question was migrated from the Slack channel #questions.
Original Question:
Hi everyone, I’m exploring the Flower framework for federated learning using this example, where each client has its own training and validation sets. I’m interested in implementing leave-one-out cross-validation (LOOCV) with this setup. Is it possible to configure it so that all clients hold their own training set, while one client at a time is used for evaluation, allowing each client a turn for validation? I would appreciate any insights or suggestions!
Answer:
Yes, you can partition the dataset to split the dataset to clients, check out Flower Datasets 0.0.2 we have popular datasets ready to be used in a federated setting.
To sample evaluation for one client only, you can change the server code during evaluate round, so that you only send evaluate message to one client.