flwr.client.mod.centraldp_mods의 소스 코드

# Copyright 2024 Flower Labs GmbH. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Clipping modifiers for central DP with client-side clipping."""


from logging import INFO

from flwr.client.typing import ClientAppCallable
from flwr.common import ndarrays_to_parameters, parameters_to_ndarrays
from flwr.common import recordset_compat as compat
from flwr.common.constant import MessageType
from flwr.common.context import Context
from flwr.common.differential_privacy import (
    compute_adaptive_clip_model_update,
    compute_clip_model_update,
)
from flwr.common.differential_privacy_constants import KEY_CLIPPING_NORM, KEY_NORM_BIT
from flwr.common.logger import log
from flwr.common.message import Message




def fixedclipping_mod(
    msg: Message, ctxt: Context, call_next: ClientAppCallable
) -> Message:
    """Client-side fixed clipping modifier.

    This mod needs to be used with the DifferentialPrivacyClientSideFixedClipping
    server-side strategy wrapper.

    The wrapper sends the clipping_norm value to the client.

    This mod clips the client model updates before sending them to the server.

    It operates on messages of type `MessageType.TRAIN`.

    Notes
    -----
    Consider the order of mods when using multiple.

    Typically, fixedclipping_mod should be the last to operate on params.
    """
    if msg.metadata.message_type != MessageType.TRAIN:
        return call_next(msg, ctxt)
    fit_ins = compat.recordset_to_fitins(msg.content, keep_input=True)
    if KEY_CLIPPING_NORM not in fit_ins.config:
        raise KeyError(
            f"The {KEY_CLIPPING_NORM} value is not supplied by the "
            f"DifferentialPrivacyClientSideFixedClipping wrapper at"
            f" the server side."
        )

    clipping_norm = float(fit_ins.config[KEY_CLIPPING_NORM])
    server_to_client_params = parameters_to_ndarrays(fit_ins.parameters)

    # Call inner app
    out_msg = call_next(msg, ctxt)

    # Check if the msg has error
    if out_msg.has_error():
        return out_msg

    fit_res = compat.recordset_to_fitres(out_msg.content, keep_input=True)

    client_to_server_params = parameters_to_ndarrays(fit_res.parameters)

    # Clip the client update
    compute_clip_model_update(
        client_to_server_params,
        server_to_client_params,
        clipping_norm,
    )

    log(
        INFO, "fixedclipping_mod: parameters are clipped by value: %.4f.", clipping_norm
    )

    fit_res.parameters = ndarrays_to_parameters(client_to_server_params)
    out_msg.content = compat.fitres_to_recordset(fit_res, keep_input=True)
    return out_msg





def adaptiveclipping_mod(
    msg: Message, ctxt: Context, call_next: ClientAppCallable
) -> Message:
    """Client-side adaptive clipping modifier.

    This mod needs to be used with the DifferentialPrivacyClientSideAdaptiveClipping
    server-side strategy wrapper.

    The wrapper sends the clipping_norm value to the client.

    This mod clips the client model updates before sending them to the server.

    It also sends KEY_NORM_BIT to the server for computing the new clipping value.

    It operates on messages of type `MessageType.TRAIN`.

    Notes
    -----
    Consider the order of mods when using multiple.

    Typically, adaptiveclipping_mod should be the last to operate on params.
    """
    if msg.metadata.message_type != MessageType.TRAIN:
        return call_next(msg, ctxt)

    fit_ins = compat.recordset_to_fitins(msg.content, keep_input=True)

    if KEY_CLIPPING_NORM not in fit_ins.config:
        raise KeyError(
            f"The {KEY_CLIPPING_NORM} value is not supplied by the "
            f"DifferentialPrivacyClientSideFixedClipping wrapper at"
            f" the server side."
        )
    if not isinstance(fit_ins.config[KEY_CLIPPING_NORM], float):
        raise ValueError(f"{KEY_CLIPPING_NORM} should be a float value.")
    clipping_norm = float(fit_ins.config[KEY_CLIPPING_NORM])
    server_to_client_params = parameters_to_ndarrays(fit_ins.parameters)

    # Call inner app
    out_msg = call_next(msg, ctxt)

    # Check if the msg has error
    if out_msg.has_error():
        return out_msg

    fit_res = compat.recordset_to_fitres(out_msg.content, keep_input=True)

    client_to_server_params = parameters_to_ndarrays(fit_res.parameters)

    # Clip the client update
    norm_bit = compute_adaptive_clip_model_update(
        client_to_server_params,
        server_to_client_params,
        clipping_norm,
    )
    log(
        INFO,
        "adaptiveclipping_mod: parameters are clipped by value: %.4f.",
        clipping_norm,
    )

    fit_res.parameters = ndarrays_to_parameters(client_to_server_params)

    fit_res.metrics[KEY_NORM_BIT] = norm_bit
    out_msg.content = compat.fitres_to_recordset(fit_res, keep_input=True)
    return out_msg