Evaluation

There are two main approaches to evaluate models in federated learning systems: centralized (or server-side) evaluation and federated (or client-side) evaluation.

Centralized Evaluation

Built-In Strategies

All built-in strategies support centalized evaluation by providing an evaluation function during initialization. An evaluation function is any function that can take the current global model parameters as input and return evaluation results:

def get_eval_fn(model):
    """Return an evaluation function for server-side evaluation."""

    # Load data and model here to avoid the overhead of doing it in `evaluate` itself
    (x_train, y_train), _ = tf.keras.datasets.cifar10.load_data()

    # Use the last 5k training examples as a validation set
    x_val, y_val = x_train[45000:50000], y_train[45000:50000]

    # The `evaluate` function will be called after every round
    def evaluate(weights: fl.common.Weights) -> Optional[Tuple[float, float]]:
        model.set_weights(weights)  # Update model with the latest parameters
        loss, accuracy = model.evaluate(x_val, y_val)
        return loss, accuracy

    return evaluate

# Load and compile model for server-side parameter evaluation
model = tf.keras.applications.EfficientNetB0(
    input_shape=(32, 32, 3), weights=None, classes=10
)
model.compile("adam", "sparse_categorical_crossentropy", metrics=["accuracy"])


# Create strategy
strategy = fl.server.strategy.FedAvg(
    # ... other FedAvg arguments
    eval_fn=get_eval_fn(model),
)

# Start Flower server for four rounds of federated learning
fl.server.start_server("[::]:8080", strategy=strategy)

Custom Strategies

The Strategy abstraction provides a method called evaluate that can direcly be used to evaluate the current global model parameters. The current server implementation calls evaluate after parameter aggregation and before federated evaluation (see next paragraph).

Federated Evaluation

Implementing Federated Evaluation

Client-side evaluation happens in the Client.evaluate method and can be configured from the server side.

class CifarClient(fl.client.NumPyClient):
    def __init__(self, model, x_train, y_train, x_test, y_test):
        self.model = model
        self.x_train, self.y_train = x_train, y_train
        self.x_test, self.y_test = x_test, y_test

    def get_parameters(self):
        # ...

    def fit(self, parameters, config):
        # ...

    def evaluate(self, parameters, config):
        """Evaluate parameters on the locally held test set."""

        # Update local model with global parameters
        self.model.set_weights(parameters)

        # Get config values
        steps: int = config["val_steps"]

        # Evaluate global model parameters on the local test data and return results
        loss, accuracy = self.model.evaluate(self.x_test, self.y_test, 32, steps=steps)
        num_examples_test = len(self.x_test)
        return loss, num_examples_test, {"accuracy": accuracy}

Configuring Federated Evaluation

Federated evaluation can be configured from the server side. Built-in strategies support the following arguments:

  • fraction_eval: a float defining the fraction of clients that will be selected for evaluation. If fraction_eval is set to 0.1 and 100 clients are connected to the server, then 10 will be randomly selected for evaluation.

  • min_eval_clients: an int: the minimum number of clients to be selected for evaluation. If fraction_eval is set to 0.1, min_eval_clients is set to 20, and 100 clients are connected to the server, then 20 clients will be selected for evaluation.

  • min_available_clients: an int that defines the minimum number of clients which need to be connected to the server before a round of federated evaluation can start. If fewer than min_available_clients are connected to the server, the server will wait until more clients are connected before it continues to sample clients for evaluation.

  • on_evaluate_config_fn: a function that returns a configuration dictionary which will be sent to the selected clients. The function will be called during each round and provides a convenient way to customize client-side evaluation from the server side, for example, to configure the number of validation steps performed.

def evaluate_config(rnd: int):
    """Return evaluation configuration dict for each round.
    Perform five local evaluation steps on each client (i.e., use five
    batches) during rounds one to three, then increase to ten local
    evaluation steps.
    """
    val_steps = 5 if rnd < 4 else 10
    return {"val_steps": val_steps}

# Create strategy
strategy = fl.server.strategy.FedAvg(
    # ... other FedAvg agruments
    fraction_eval=0.2,
    min_eval_clients=2,
    min_available_clients=10,
    on_evaluate_config_fn=evaluate_config,
)

# Start Flower server for four rounds of federated learning
fl.server.start_server("[::]:8080", strategy=strategy)

Evaluating Local Model Updates During Training

Model parameters can also be evaluated during training. Client.fit can return arbitrary evaluation results as a dictionary:

class CifarClient(fl.client.NumPyClient):
    def __init__(self, model, x_train, y_train, x_test, y_test):
        self.model = model
        self.x_train, self.y_train = x_train, y_train
        self.x_test, self.y_test = x_test, y_test

    def get_parameters(self):
        # ...

    def fit(self, parameters, config):
        """Train parameters on the locally held training set."""

        # Update local model parameters
        self.model.set_weights(parameters)

        # Train the model using hyperparameters from config
        history = self.model.fit(
            self.x_train, self.y_train, batch_size=32, epochs=2, validation_split=0.1
        )

        # Return updated model parameters and validation results
        parameters_prime = self.model.get_weights()
        num_examples_train = len(self.x_train)
        results = {
            "loss": history.history["loss"][0],
            "accuracy": history.history["accuracy"][0],
            "val_loss": history.history["val_loss"][0],
            "val_accuracy": history.history["val_accuracy"][0],
        }
        return parameters_prime, num_examples_train, results

    def evaluate(self, parameters, config):
        # ...

Full Code Example

For a full code example that uses both centralized and federated evaluation, see the Advanced TensorFlow Example (the same approach can be applied to workloads implemented in any other framework): https://github.com/adap/flower/tree/main/examples/advanced_tensorflow