Eth Account Strategy

exception rizemind.authentication.eth_account_strategy.CannotRecoverSignerException

Bases: RizemindException

The signer of a model update could not be recovered.

exception rizemind.authentication.eth_account_strategy.CannotTrainException(address: str)

Bases: RizemindException

An attempt was made to train with an unauthorized address.

class rizemind.authentication.eth_account_strategy.EthAccountStrategy(strat: Strategy, swarm: SupportsEthAccountStrategy, account: BaseAccount)

Bases: Strategy

A federated learning strategy that verifies model authenticity.

This strategy wraps an existing Flower Strategy to ensure that only authorized clients can contribute training updates. It verifies cryptographic signatures against a blockchain-based model registry. If a client is not authorized, it is added to the failures list with a CannotTrainException.

strat

The base Flower Strategy to wrap.

Type:

flwr.server.strategy.strategy.Strategy

swarm

The blockchain-based model registry.

Type:

rizemind.authentication.typing.SupportsEthAccountStrategy

address

The contract address of the swarm.

Type:

str

account

The Ethereum account used for signing.

Type:

eth_account.signers.base.BaseAccount

Example Usage:

strategy = SomeBaseStrategy()

model_registry = SwarmV1.from_address(address=”0xMY_MODEL_ADDRESS”)

eth_strategy = EthAccountStrategy(strategy, model_registry)

account: BaseAccount

address: str

aggregate_evaluate(server_round, results, failures)

Aggregate evaluation results by delegating to the wrapped strategy.

Parameters:

• server_round – The current server round.

• results – A list of tuples (ClientProxy, EvaluateRes) received from clients.

• failures – A list of failures encountered during evaluation.

Returns:

The aggregated evaluation result as returned by the wrapped strategy’s aggregate_evaluate.

aggregate_fit(server_round, results, failures)

Aggregate fit results from authorized clients only.

Recovers the signer address from each client’s fit result, tags the client with the recovered address, and filters out contributions from non-authorized addresses. Unauthorized attempts are recorded as failures. Delegates final aggregation to the wrapped strategy.

Parameters:

• server_round – The current server round.

• results – A list of tuples (ClientProxy, FitRes) received from clients.

• failures – A list that will be extended with failures that occur during processing.

Returns:

The aggregated result as returned by the wrapped strategy’s aggregate_fit.

configure_evaluate(server_round, parameters, client_manager)

Prepare evaluation instructions using authenticated client manager.

Wraps the base strategy’s configure_evaluate with AuthenticatedClientManager and delegates to the wrapped strategy.

Parameters:

• server_round – The current server round.

• parameters – The global model parameters to send to clients.

• client_manager – The Flower client manager.

Returns:

The list of client evaluation instructions produced by the wrapped strategy.

configure_fit(server_round, parameters, client_manager)

Prepare fit instructions and attach notary metadata.

Wraps the base strategy’s configure_fit to use AuthenticatedClientManager and appends a signed notary payload to each client’s config so that clients can sign their updates.

Parameters:

• server_round – The current server round.

• parameters – The global model parameters to send to clients.

• client_manager – The Flower client manager.

Returns:

The list of client instructions produced by the wrapped strategy with notary metadata attached to each instruction’s config.

evaluate(server_round, parameters)

Evaluate the current global model via the wrapped strategy.

Parameters:

• server_round – The current server round.

• parameters – The global model parameters to evaluate.

Returns:

The evaluation result as returned by the wrapped strategy’s evaluate.

initialize_parameters(client_manager)

Initializes model parameters.

strat: Strategy

swarm: SupportsEthAccountStrategy