Client

This page documents the client-side components of ICOS-FL, particularly the classes and functions defined in icos_fl.client.client.

Client Components

class icos_fl.client.client.IcosClient(client_id, client_state, model, trainloader, valloader, local_epochs, metric)

Client implementation for ICOS FL with time series prediction.

This client trains an LSTM model for time series prediction in a federated learning setting. It handles local model training, evaluation, and parameter management as part of the federated learning process.

Parameters:

  • client_id (Union[int, str]) – Unique identifier for this client

  • client_state (RecordSet) – RecordSet to maintain state across rounds

  • model (LSTMModel) – The LSTM model to train

  • trainloader (torch.utils.data.DataLoader) – DataLoader for training data

  • valloader (torch.utils.data.DataLoader) – DataLoader for validation data

  • local_epochs (int) – Number of local training epochs per FL round

  • metric (str) – Name of the metric being predicted

get_parameters(config)

Get model parameters as a list of NumPy arrays.

Parameters:

config (Dict[str, Any]) – Configuration parameters

Returns:

List of model parameter arrays

Return type:

List[NDArrays]

set_parameters(parameters)

Set model parameters from a list of NumPy arrays.

Parameters:

parameters (List[NDArrays]) – List of model parameter arrays

fit(parameters, config)

Train the model on the local dataset.

Parameters:

  • parameters (List[NDArrays]) – Current global model parameters

  • config (Dict[str, Any]) – Configuration parameters for training

Returns:

Tuple containing (updated model parameters, number of training samples, metrics)

Return type:

Tuple[List[NDArrays], int, Dict[str, Any]]

evaluate(parameters, config)

Evaluate the model on the local validation dataset.

Parameters:

  • parameters (List[NDArrays]) – Current global model parameters

  • config (Dict[str, Any]) – Configuration parameters for evaluation

Returns:

Tuple containing (loss, number of validation samples, metrics)

Return type:

Tuple[float, int, Dict[str, Any]]

icos_fl.client.client.client_fn(context)

Create and return a Flower client instance.

This function is used by the ClientApp to create a client instance for each node in the federated learning system. It extracts configuration from the context, initializes the model, and prepares data loaders for training and evaluation.

Parameters:

context (Context) – Flower client context with configuration and node info

Returns:

Instantiated NumPyClient

Return type:

NumPyClient

ClientApp Definition

icos_fl.client.client.app

The Flower ClientApp instance used for federated learning.

This is the main entry point for the client-side federated learning process. It uses the client_fn function to create client instances.

Type:

ClientApp

Client Workflow

The client workflow in ICOS-FL follows these steps:

  1. Initialization: The client is initialized with a model, data loaders, and configuration

  2. Data Fetching: The client fetches local system metrics data from DataClay

  3. Model Parameter Exchange: The client receives the global model parameters

  4. Local Training: The client trains the model on local data for a specified number of epochs

  5. Local Evaluation: The client evaluates the model on a validation set

  6. Parameter Update: The client sends the updated parameters back to the server

Example Usage

# Run the client with Flower CLI
$ flwr run . remote-deployment

# Or programmatically
from icos_fl.client.client import app

app.start(server_address="127.0.0.1:9092")