PFLlib/system/flcore/clients/clientapple.py

# PFLlib: Personalized Federated Learning Algorithm Library
# Copyright (C) 2021  Jianqing Zhang

# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.

# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.

# You should have received a copy of the GNU General Public License along
# with this program; if not, write to the Free Software Foundation, Inc.,
# 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.

import copy
import math
import torch
import numpy as np
import time
from flcore.clients.clientbase import Client


class clientAPPLE(Client):
    def __init__(self, args, id, train_samples, test_samples, **kwargs):
        super().__init__(args, id, train_samples, test_samples, **kwargs)

        self.drlr = args.dr_learning_rate
        self.num_clients = args.num_clients
        self.lamda = 1
        self.mu = args.mu
        self.L = int(args.L * args.global_rounds)
        self.learning_rate = self.learning_rate * self.num_clients

        self.model_cs = []

        self.ps = [1/args.num_clients for _ in range(self.num_clients)]
        self.p0 = None
        self.model_c = copy.deepcopy(self.model)

    def train(self, R):
        trainloader = self.load_train_data()
        
        start_time = time.time()

        # self.model.to(self.device)
        self.model.train()

        max_local_epochs = self.local_epochs
        if self.train_slow:
            max_local_epochs = np.random.randint(1, max_local_epochs // 2)

        for epoch in range(max_local_epochs):
            for i, (x, y) in enumerate(trainloader):
                if type(x) == type([]):
                    x[0] = x[0].to(self.device)
                else:
                    x = x.to(self.device)
                y = y.to(self.device)
                if self.train_slow:
                    time.sleep(0.1 * np.abs(np.random.rand()))

                self.aggregate_parameters()

                output = self.model(x)
                loss = self.loss(output, y)
                self.optimizer.zero_grad()
                loss.backward()

                for param_c, param in zip(self.model_cs[self.id].parameters(), self.model.parameters()):
                    param_c.data = param_c - self.learning_rate * param.grad * self.ps[self.id]

                for cid in range(self.num_clients):
                    cnt = 0
                    p_grad = 0
                    for param_c, param in zip(self.model_cs[cid].parameters(), self.model.parameters()):
                        p_grad += torch.mean(param.grad * param_c).item()
                        cnt += 1
                    p_grad = p_grad / cnt
                    p_grad = p_grad + self.lamda * self.mu * (self.ps[cid] - self.p0[cid])
                    self.ps[cid] = self.ps[cid] - self.drlr * p_grad

        if R < self.L:
            self.lamda = (math.cos(R * math.pi / self.L) + 1) / 2
        else:
            self.lamda = 0

        # recover self.model_cs[self.id] for other clients
        for param_c, param_ in zip(self.model_cs[self.id].parameters(), self.model_c.parameters()):
            param_c.data = param_.data.clone()

        self.model_c = copy.deepcopy(self.model)

        if self.learning_rate_decay:
            self.learning_rate_scheduler.step()

        self.train_time_cost['num_rounds'] += 1
        self.train_time_cost['total_cost'] += time.time() - start_time

        
    def set_models(self, model_cs):
        self.model_cs = model_cs

    def add_parameters(self, w, client_model):
        for server_param, client_param in zip(self.model.parameters(), client_model.parameters()):
            server_param.data += client_param.data.clone() * w

    def aggregate_parameters(self):
        assert (len(self.model_cs) > 0)

        for param in self.model.parameters():
            param.data = torch.zeros_like(param.data)
            
        for w, client_model in zip(self.ps, self.model_cs):
            self.add_parameters(w, client_model)