ClientApp

class ClientApp(client_fn: Callable[[Context], Client] | None = None, mods: list[Callable[[Message, Context, Callable[[Message, Context], Message]], Message]] | None = None)

Bases: object

Flower ClientApp.

Examples

Assuming a typical Client implementation named FlowerClient, you can wrap it in a ClientApp as follows:

class FlowerClient(NumPyClient):
    # ...

def client_fn(context: Context):
   return FlowerClient().to_client()

app = ClientApp(client_fn)

Methods

evaluate([action, mods])	Register an evaluate function with the ClientApp.
lifespan()	Return a decorator that registers the lifespan fn with the client app.
query([action, mods])	Register a query function with the ClientApp.
train([action, mods])	Register a train function with the ClientApp.

evaluate(action: str = 'default', *, mods: list[Callable[[Message, Context, Callable[[Message, Context], Message]], Message]] | None = None) → Callable[[Callable[[Message, Context], Message]], Callable[[Message, Context], Message]]

Register an evaluate function with the ClientApp.

Parameters:

• action (str (default: "default")) – The action name used to route messages. Defaults to “default”.

• mods (Optional[list[Mod]] (default: None)) – A list of function-specific modifiers.

Returns:

A decorator that registers an evaluate function with the ClientApp.

Return type:

Callable[[ClientAppCallable], ClientAppCallable]

Examples

Registering an evaluate function:

app = ClientApp()

@app.evaluate()
def evaluate(message: Message, context: Context) -> Message:
    print("Executing default evaluate function")
    # Create and return an echo reply message
    return Message(message.content, reply_to=message)

Registering an evaluate function with a custom action name:

app = ClientApp()

# Messages with `message_type="evaluate.custom_action"` will be
# routed to this function.
@app.evaluate("custom_action")
def custom_action(message: Message, context: Context) -> Message:
    print("Executing evaluate function for custom action")
    return Message(message.content, reply_to=message)

Registering an evaluate function with a function-specific Flower Mod:

from flwr.client.mod import message_size_mod

app = ClientApp()

# Using the `mods` argument to apply a function-specific mod.
@app.evaluate(mods=[message_size_mod])
def evaluate(message: Message, context: Context) -> Message:
    print("Executing evaluate function with message size mod")
    # Create and return an echo reply message
    return Message(message.content, reply_to=message)

lifespan() → Callable[[Callable[[Context], Iterator[None]]], Callable[[Context], Iterator[None]]]

Return a decorator that registers the lifespan fn with the client app.

The decorated function should accept a Context object and use yield to define enter and exit behavior.

Examples

app = ClientApp()

@app.lifespan()
def lifespan(context: Context) -> None:
    # Perform initialization tasks before the app starts
    print("Initializing ClientApp")

    yield  # ClientApp is running

    # Perform cleanup tasks after the app stops
    print("Cleaning up ClientApp")

query(action: str = 'default', *, mods: list[Callable[[Message, Context, Callable[[Message, Context], Message]], Message]] | None = None) → Callable[[Callable[[Message, Context], Message]], Callable[[Message, Context], Message]]

Register a query function with the ClientApp.

Parameters:

• action (str (default: "default")) – The action name used to route messages. Defaults to “default”.

• mods (Optional[list[Mod]] (default: None)) – A list of function-specific modifiers.

Returns:

A decorator that registers a query function with the ClientApp.

Return type:

Callable[[ClientAppCallable], ClientAppCallable]

Examples

Registering a query function:

app = ClientApp()

@app.query()
def query(message: Message, context: Context) -> Message:
    print("Executing default query function")
    # Create and return an echo reply message
    return Message(message.content, reply_to=message)

Registering a query function with a custom action name:

app = ClientApp()

# Messages with `message_type="query.custom_action"` will be
# routed to this function.
@app.query("custom_action")
def custom_action(message: Message, context: Context) -> Message:
    print("Executing query function for custom action")
    return Message(message.content, reply_to=message)

Registering a query function with a function-specific Flower Mod:

from flwr.client.mod import message_size_mod

app = ClientApp()

# Using the `mods` argument to apply a function-specific mod.
@app.query(mods=[message_size_mod])
def query(message: Message, context: Context) -> Message:
    print("Executing query function with message size mod")
    # Create and return an echo reply message
    return Message(message.content, reply_to=message)

train(action: str = 'default', *, mods: list[Callable[[Message, Context, Callable[[Message, Context], Message]], Message]] | None = None) → Callable[[Callable[[Message, Context], Message]], Callable[[Message, Context], Message]]

Register a train function with the ClientApp.

Parameters:

• action (str (default: "default")) – The action name used to route messages. Defaults to “default”.

• mods (Optional[list[Mod]] (default: None)) – A list of function-specific modifiers.

Returns:

A decorator that registers a train function with the ClientApp.

Return type:

Callable[[ClientAppCallable], ClientAppCallable]

Examples

Registering a train function:

app = ClientApp()

@app.train()
def train(message: Message, context: Context) -> Message:
    print("Executing default train function")
    # Create and return an echo reply message
    return Message(message.content, reply_to=message)

Registering a train function with a custom action name:

app = ClientApp()

# Messages with `message_type="train.custom_action"` will be
# routed to this function.
@app.train("custom_action")
def custom_action(message: Message, context: Context) -> Message:
    print("Executing train function for custom action")
    return Message(message.content, reply_to=message)

Registering a train function with a function-specific Flower Mod:

from flwr.client.mod import message_size_mod

app = ClientApp()

# Using the `mods` argument to apply a function-specific mod.
@app.train(mods=[message_size_mod])
def train(message: Message, context: Context) -> Message:
    print("Executing train function with message size mod")
    # Create and return an echo reply message
    return Message(message.content, reply_to=message)