-
Notifications
You must be signed in to change notification settings - Fork 27
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
example/implementation for FedBalancer, with a new datasampler category
Add a new category, "datasampler", which selects trainers' data at FL rounds. Add FedBalancer (Jaemin Shin et al., FedBalancer: Data and Pace Control for Efficient Federated Learning on Heterogeneous Clients, MobiSys'22) as a new datasampler, which actively selects more important training samples of trainers to speed up global FL. Implement a control scheme of "deadline", which is only used for fedbalancer's sample selection at this version. Deadline-based round termination will be supported in later updates. Refer to lib/python/flame/examples/fedbalancer_mnist/ for example of running fedbalancer Things that current version of fedbalancer do not support: - Advanced trainer selection with Oort proposed in FedBalancer - Other FL modes: hybrid, hierarchical
- Loading branch information
1 parent
c71ffc0
commit adc2b06
Showing
20 changed files
with
1,554 additions
and
30 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,78 @@ | ||
| # Copyright 2023 Cisco Systems, Inc. and its affiliates | ||
| # | ||
| # 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. | ||
| # | ||
| # SPDX-License-Identifier: Apache-2.0 | ||
| """datasampler abstract class.""" | ||
|
|
||
| from abc import ABC, abstractmethod | ||
| from typing import Any | ||
|
|
||
| from flame.channel import Channel | ||
|
|
||
|
|
||
| class AbstractDataSampler(ABC): | ||
| class AbstractTrainerDataSampler(ABC): | ||
| """Abstract base class for trainer-side datasampler implementation.""" | ||
|
|
||
| def __init__(self, **kwargs) -> None: | ||
| """Initialize an instance with keyword-based arguments.""" | ||
| for key, value in kwargs.items(): | ||
| setattr(self, key, value) | ||
|
|
||
| @abstractmethod | ||
| def sample(self, dataset: Any, **kwargs) -> Any: | ||
| """Abstract method to sample data. | ||
| Parameters | ||
| ---------- | ||
| dataset: Dataset of a trainer to select samples from | ||
| kwargs: other arguments specific to each datasampler algorithm | ||
| Returns | ||
| ------- | ||
| dataset: Dataset that only contains selected samples | ||
| """ | ||
|
|
||
| @abstractmethod | ||
| def load_dataset(self, dataset: Any) -> Any: | ||
| """Process dataset instance for datasampler.""" | ||
|
|
||
| @abstractmethod | ||
| def get_metadata(self) -> dict[str, Any]: | ||
| """Return metadata to send to aggregator-side datasampler.""" | ||
|
|
||
| @abstractmethod | ||
| def handle_metadata_from_aggregator(self, metadata: dict[str, Any]) -> None: | ||
| """Handle aggregator metadata for datasampler.""" | ||
|
|
||
| class AbstractAggregatorDataSampler(ABC): | ||
| """Abstract base class for aggregator-side datasampler implementation.""" | ||
|
|
||
| def __init__(self, **kwargs) -> None: | ||
| """Initialize an instance with keyword-based arguments.""" | ||
| for key, value in kwargs.items(): | ||
| setattr(self, key, value) | ||
|
|
||
| @abstractmethod | ||
| def get_metadata(self, end: str, round: int) -> dict[str, Any]: | ||
| """Return metadata to send to trainer-side datasampler.""" | ||
|
|
||
| @abstractmethod | ||
| def handle_metadata_from_trainer( | ||
| self, | ||
| metadata: dict[str, Any], | ||
| end: str, | ||
| channel: Channel, | ||
| ) -> None: | ||
| """Handle trainer metadata for datasampler.""" |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,77 @@ | ||
| # Copyright 2023 Cisco Systems, Inc. and its affiliates | ||
| # | ||
| # 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. | ||
| # | ||
| # SPDX-License-Identifier: Apache-2.0 | ||
| """DefaultDataSampler class.""" | ||
|
|
||
| import logging | ||
| from typing import Any | ||
|
|
||
| from flame.channel import Channel | ||
| from flame.datasampler import AbstractDataSampler | ||
|
|
||
| logger = logging.getLogger(__name__) | ||
|
|
||
|
|
||
| class DefaultDataSampler(AbstractDataSampler): | ||
| def __init__(self) -> None: | ||
| self.trainer_data_sampler = DefaultDataSampler.DefaultTrainerDataSampler() | ||
| self.aggregator_data_sampler = DefaultDataSampler.DefaultAggregatorDataSampler() | ||
|
|
||
| class DefaultTrainerDataSampler(AbstractDataSampler.AbstractTrainerDataSampler): | ||
| """A default trainer-side datasampler class.""" | ||
|
|
||
| def __init__(self, **kwargs): | ||
| """Initailize instance.""" | ||
| super().__init__() | ||
|
|
||
| def sample(self, dataset: Any, **kwargs) -> Any: | ||
| """Return all dataset from the given dataset.""" | ||
| logger.debug("calling default datasampler") | ||
|
|
||
| return dataset | ||
|
|
||
| def load_dataset(self, dataset: Any) -> None: | ||
| """Change dataset instance to return index with each sample.""" | ||
| return dataset | ||
|
|
||
| def get_metadata(self) -> dict[str, Any]: | ||
| """Return metadata to send to aggregator-side datasampler.""" | ||
| return {} | ||
|
|
||
| def handle_metadata_from_aggregator(self, metadata: dict[str, Any]) -> None: | ||
| """Handle aggregator metadata for datasampler.""" | ||
| pass | ||
|
|
||
| class DefaultAggregatorDataSampler( | ||
| AbstractDataSampler.AbstractAggregatorDataSampler | ||
| ): | ||
| """A default aggregator-side datasampler class.""" | ||
|
|
||
| def __init__(self, **kwargs): | ||
| """Initailize instance.""" | ||
| super().__init__() | ||
|
|
||
| def get_metadata(self, end: str, round: int) -> dict[str, Any]: | ||
| """Return metadata to send to trainer-side datasampler.""" | ||
| return {} | ||
|
|
||
| def handle_metadata_from_trainer( | ||
| self, | ||
| metadata: dict[str, Any], | ||
| end: str, | ||
| channel: Channel, | ||
| ) -> None: | ||
| """Handle trainer metadata for datasampler.""" | ||
| pass |
Oops, something went wrong.