V2.0

[felicityfelicity]

from ..base.baseAggregator import ServerAggregator
import numpy as np

class CCVRAggregator(ServerAggregator):
def init(self):
"""
初始化 CCVR Aggregator
"""
super().init()

def _on_before_aggregation(self):
    """
    聚合前的操作（可选）
    """
    pass

def _on_after_aggregation(self):
    """
    聚合后的操作（可选）
    """
    pass

def _aggregate_alg(self, client_statistics_list=None):
    """
    实现 CCVR 聚合算法
    - client_statistics_list: 客户端上传的统计信息列表
        [
            {class_label: (mu, sigma, n_samples), ...},
            ...
        ]
    """
    if client_statistics_list is None:
        client_statistics_list = self.model_pool

    global_statistics = {}
    for client_stats in client_statistics_list:
        for class_label, (mu, sigma, n_samples) in client_stats.items():
            if class_label not in global_statistics:
                global_statistics[class_label] = {
                    "mu_sum": np.zeros_like(mu),
                    "sigma_sum": np.zeros_like(sigma),
                    "n_samples": 0
                }

            global_statistics[class_label]["mu_sum"] += mu * n_samples
            global_statistics[class_label]["sigma_sum"] += sigma * n_samples
            global_statistics[class_label]["n_samples"] += n_samples

    # 计算全局均值和协方差
    aggregated_statistics = {}
    for class_label, stats in global_statistics.items():
        n_total = stats["n_samples"]
        if n_total > 0:
            mu_global = stats["mu_sum"] / n_total
            sigma_global = stats["sigma_sum"] / n_total
            aggregated_statistics[class_label] = (mu_global, sigma_global)

    return aggregated_statistics

if name == 'main':
# 示例：服务器接收到的客户端统计信息
client_statistics_list = [
{
0: (np.array([mu_0_c1]), np.array([sigma_0_c1]), n_0_c1),
1: (np.array([mu_1_c1]), np.array([sigma_1_c1]), n_1_c1),
},
{
0: (np.array([mu_0_c2]), np.array([sigma_0_c2]), n_0_c2),
1: (np.array([mu_1_c2]), np.array([sigma_1_c2]), n_1_c2),
},
]

aggregator = CCVRAggregator()
aggregated_statistics = aggregator._aggregate_alg(client_statistics_list)

# 输出示例
for class_label, (mu, sigma) in aggregated_statistics.items():
    print(f"Class {class_label}:")
    print(f"  Global Mu: {mu}")
    print(f"  Global Sigma: {sigma}")