-
Notifications
You must be signed in to change notification settings - Fork 4
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
13 changed files
with
351 additions
and
66 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 +1,147 @@ | ||
| # Copyright 2022 iFLYTEK. 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. | ||
| # ============================================================================== | ||
|
|
||
| import sys | ||
| import pickle | ||
| import numpy as np | ||
| from loguru import logger | ||
| from typing import Dict, Any, List | ||
| from iflearner.communication.homo import homo_pb2 | ||
| from iflearner.business.homo.strategy.strategy_client import StrategyClient | ||
|
|
||
|
|
||
| class STCClient(StrategyClient): | ||
| """Implement the STC method base on the paper (https://ieeexplore.ieee.org/document/8889996). | ||
| """ | ||
|
|
||
| def __init__(self) -> None: | ||
| super().__init__() | ||
|
|
||
| self._top_fraction: float = 0.1 | ||
| self._weights: Dict = None | ||
| self._residuals: Dict = None | ||
| self._enable_residuals: bool = True | ||
| logger.info( | ||
| f"STC client, top fraction: {self._top_fraction}, enable residuals: {self._enable_residuals}") | ||
|
|
||
| def generate_upload_param(self, epoch: int, data: Dict[Any, Any], metrics: Dict[str, float] = None) -> homo_pb2.UploadParam: | ||
| if self._residuals is None: | ||
| self._residuals = {} | ||
| self._weights = {} | ||
| for k, v in data.items(): | ||
| self._residuals[k] = np.zeros(v.size) | ||
| self._weights[k] = np.zeros(v.size) | ||
|
|
||
| compressed_data = {} | ||
| for k, v in data.items(): | ||
| compressed_data[k] = homo_pb2.Parameter(shape=v.shape) | ||
|
|
||
| ravel_v = v.ravel() | ||
|
|
||
| weight_difference = ravel_v - self._weights[k] | ||
| if self._enable_residuals: | ||
| self._residuals[k] += weight_difference | ||
|
|
||
| self._weights[k] = ravel_v | ||
|
|
||
| out = None | ||
| mean = 0.0 | ||
| if self._enable_residuals: | ||
| out, mean = self._compression(self._residuals[k]) | ||
| self._residuals[k] -= out | ||
| else: | ||
| out, mean = self._compression(weight_difference) | ||
|
|
||
| sparse_data = self._encode_sparse_array(out, mean) | ||
| compressed_data[k].custom_values = sparse_data | ||
|
|
||
| return homo_pb2.UploadParam(epoch=epoch, parameters=compressed_data, metrics=metrics) | ||
|
|
||
| def aggregate_result(self) -> homo_pb2.AggregateResult: | ||
| for k, v in self._aggregate_result_np.items(): | ||
| self._weights[k] = v.flatten() | ||
|
|
||
| return self._aggregate_result_np | ||
|
|
||
| def _compression(self, T: np.array) -> np.array: | ||
| """Compress a array. | ||
| Args: | ||
| T (np.array): The array that needs to be compressed. | ||
| Returns: | ||
| np.array: The compressed array. | ||
| """ | ||
| T_abs = np.absolute(T) | ||
| n_top = int(np.ceil(T_abs.size * self._top_fraction)) | ||
| topk = T_abs[np.argpartition(T_abs, -n_top)[-n_top:]] | ||
| mean = np.mean(topk) | ||
| min_topk = topk.min() | ||
| out_ = np.where(T >= min_topk, mean, 0.0) | ||
| out = np.where(T <= -min_topk, -mean, out_) | ||
| return out, mean | ||
|
|
||
| def _encode_sparse_array(self, arr: np.array, mean_value: np.float64) -> bytes: | ||
| """Encode a sparse array to bytes. | ||
| Args: | ||
| arr (np.array): A sparse array. | ||
| Returns: | ||
| bytes: The data that dumped by pickle. | ||
| """ | ||
| logger.info( | ||
| f"Encode a sparse array, size: {arr.size}") | ||
|
|
||
| positive_horizontal_coordinates = [] | ||
| positive_vertical_coordinates = [] | ||
| negative_horizontal_coordinates = [] | ||
| negative_vertical_coordinates = [] | ||
| horizontal_coordinate_type = np.uint8 | ||
| vertical_coordinate_type = np.uint8 | ||
|
|
||
| uint8_len = np.iinfo(np.uint8).max + 1 | ||
| uint16_len = np.iinfo(np.uint16).max + 1 | ||
| if arr.size > uint8_len * uint16_len: | ||
| horizontal_coordinate_type = np.uint16 | ||
| if arr.size > uint8_len * uint8_len: | ||
| vertical_coordinate_type = np.uint16 | ||
|
|
||
| horizontal_index = 0 | ||
| vertical_index = 0 | ||
| for item in arr: | ||
| if horizontal_index > np.iinfo(np.uint8).max: | ||
| horizontal_index = 0 | ||
| vertical_index += 1 | ||
| if item > 0: | ||
| positive_horizontal_coordinates.append( | ||
| horizontal_coordinate_type(horizontal_index)) | ||
| positive_vertical_coordinates.append( | ||
| vertical_coordinate_type(vertical_index)) | ||
| elif item < 0: | ||
| negative_horizontal_coordinates.append( | ||
| horizontal_coordinate_type(horizontal_index)) | ||
| negative_vertical_coordinates.append( | ||
| vertical_coordinate_type(vertical_index)) | ||
|
|
||
| horizontal_index += 1 | ||
|
|
||
| my_np_tuple = (mean_value, np.array(positive_horizontal_coordinates), | ||
| np.array(positive_vertical_coordinates), np.array(negative_horizontal_coordinates), np.array(negative_vertical_coordinates)) | ||
| data = pickle.dumps(my_np_tuple) | ||
| logger.info( | ||
| f"After encoding, positive coordinates num: {len(positive_horizontal_coordinates)}, negative coordinates num: {len(negative_horizontal_coordinates)}, the size is {len(data)}") | ||
|
|
||
| return data |
Oops, something went wrong.