File size: 17,012 Bytes
29d330e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 | import numpy as np
import argparse
import torch
import torch.nn as nn
import random
import math
import os
import copy
from torchvision import models
from easyfsl.methods import PrototypicalNetworks
from easyfsl.utils import evaluate
from logging import WARNING
from utils_aggregation import aggregate, aggregate_krum, aggregate_median, aggregate_qffl, aggregate_trimmed_avg, weighted_loss_avg
from Client_setup import setup_client
from utils_training import train
random_seed = 0
np.random.seed(random_seed)
torch.manual_seed(random_seed)
random.seed(random_seed)
class federated_training():
def __init__(
self,
method,
run_name,
model,
station_properties,
rounds,
epochs,
num_clients,
init_weights_pth,
weighted_loss,
embedding_diff,
embedding_diff_plus_loss,
emb_diff_to_server,
momentum_vector,
lossfunction,
env,
mode=False,
):
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
self.model = model
print(f"printing path of global model -------------------{init_weights_pth}")
self.init_weights = torch.load(init_weights_pth, map_location=self.device)
self.method = method
self.rounds = rounds
self.epochs = epochs
self.global_model = copy.deepcopy(network_)
self.global_model.load_state_dict(self.init_weights)
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
self.momentum_vector: momentum_vector
self.station_train_loaders = {}
self.station_valid_loaders = {}
self.global_model_stations = {}
self.globel_model_second = {}
self.station_properties = station_properties
self.run_name = run_name
self.clients = num_clients
self.log_files = []
self.local_model_pth = []
self.weighted_loss = weighted_loss
self.embedding_diff = embedding_diff
self.embedding_diff_plus_loss = embedding_diff_plus_loss
self.emb_diff_to_server = emb_diff_to_server
self.log_emb_diff = []
self.init_weights_list = list(self.init_weights.values())
self.server_momentum = 0.0
self.server_learning_rate = 1.0
self.mode = mode
self.lsfunction = lossfunction
self.env_list_all = env
print(f"current device is {self.device}")
def setup_clients(self, num_clients, station_properties, fed_method, run_name, station_train_loaders,
station_valid_loaders, global_model_stations, globel_model_second, log_files, local_md_pth,
log_emb_diff, env_list):
for station_number in range(0, num_clients):
self.setup_client(station_properties, station_number, fed_method, run_name, station_train_loaders,
station_valid_loaders, global_model_stations, globel_model_second, log_files,
local_md_pth, log_emb_diff, env_list[station_number])
def setup_client(self, station_properties, station_number, fed_method, run_name, station_train_loaders,
station_valid_loaders, global_model_stations, globel_model_second, log_files, local_md_pth,
log_emb_diff, env_list):
setup_client(station_properties, station_number, fed_method, run_name, station_train_loaders,
station_valid_loaders, global_model_stations, globel_model_second, log_files, local_md_pth,
log_emb_diff, env_list)
def get_parameters(self, net):
print("Im Under get_parameters")
return [val.cpu().numpy() for _, val in net.state_dict().items()]
def set_parameters(self, net, parameters):
print("Im Under set_parameters")
parameters = [param.cpu().numpy() if isinstance(param, torch.Tensor) else param for param in parameters]
parameters = [torch.from_numpy(param).to(self.device) for param in parameters]
state_dict = dict(zip(net.state_dict().keys(), parameters))
net.load_state_dict(state_dict, strict=True)
def test(self, net, remote_valid_dataset):
"""Evaluate the network on the entire test set."""
# Evaluate and save the model if it's the best
net.to(self.device)
net.eval()
validation_accuracy = evaluate(net, remote_valid_dataset, self.device, tqdm_prefix="Validation")
return validation_accuracy
def fit(self, weights_array_numexpl, server_round):
if self.method == "FedAvg":
print("-------------------------Im using FedAvg--------------------------------")
updated_weights = aggregate(weights_array_numexpl)
elif self.method == "FedAvgM":
print("-------------------------Im using FedAvgM--------------------------------")
fedavg_result = aggregate(weights_array_numexpl)
assert self.init_weights is not None, "Model needs to be initialized."
init_weights_list_device = [x.to(self.device) for x in self.init_weights_list]
fedavg_result_device = [y.to(self.device) for y in fedavg_result]
pseudo_gradient = [x - y for x, y in zip(init_weights_list_device, fedavg_result_device)]
if isinstance(self.server_momentum, float):
print("momentum is float")
else:
print("its tuple")
print({"im underb server momentum calculation-------------------------------"})
if self.server_momentum > 0.0:
if server_round > 1:
assert self.momentum_vector, "Momentum should have been created on round 1."
self.momentum_vector = [
self.server_momentum * x + y
for x, y in zip(self.momentum_vector, pseudo_gradient)
]
else:
self.momentum_vector = pseudo_gradient
# Assign pseudo-gradient as momentum
pseudo_gradient = self.momentum_vector
# SGD update
updated_weights = [
x - self.server_learning_rate * y
for x, y in zip(init_weights_list_device, pseudo_gradient)
]
# Update current weights
self.init_weights_list = init_weights_list_device
elif self.method == "FedMedian":
print("-------------------------Im using FedMedian--------------------------------")
updated_weights = aggregate_median(weights_array_numexpl)
return updated_weights
def train_fedretaed(self):
##set up for clients
self.setup_clients(self.clients,
self.station_properties,
self.method,
self.run_name,
self.station_train_loaders,
self.station_valid_loaders,
self.global_model_stations,
self.globel_model_second,
self.log_files,
self.local_model_pth,
self.log_emb_diff,
self.env_list_all)
prev_round_diff = [0] * self.clients ##added
weights_epoch_diff = [0] * self.clients ##added
###########################early-epoch based methods starts here################################
for rn in range(self.rounds):
print(f"im runnning for round :{rn + 1}")
print(f"weights_epoch_diff for round :{rn + 1} are---- {weights_epoch_diff}")
# (math.floor(max(diff_2)/max(diff)*7))
if rn == 1:
eps = self.epochs - 1
for i in range(len(prev_round_diff)):
prev_round_diff[i] = weights_epoch_diff[i]
elif rn > 1:
eps = max((math.floor(max(weights_epoch_diff) / max(prev_round_diff) * eps)), 2)
eps = min(self.epochs, eps)
eps = eps - 1
for i in range(len(prev_round_diff)):
prev_round_diff[i] = weights_epoch_diff[i]
else:
eps = self.epochs - 1 # for rn ==0
new_epoch = [eps]*self.clients
if rn > 0:
weights = [x / max(weights_epoch_diff) for x in weights_epoch_diff]
new_epoch = [math.floor(x * y) for x, y in zip(new_epoch, weights)]
new_epoch = [x + 1 for x in new_epoch]
print(f"epochs for round :{rn + 1} are---- {new_epoch}")
#################################################################################################################
s_m = [120, 160]
s_g = 0.1
lr = 1e-2
results_fed = []
start_emb_df_accumulated = [0] * self.clients
for station_number, model in enumerate(self.global_model_stations.values(), start=0):
layer_diff_list = []
if self.mode:
new_epoch = new_epoch # ---------------------------------------added
else:
new_epoch = [self.epochs, self.epochs, self.epochs, self.epochs]
if self.emb_diff_to_server:
emb_diff_epoch = train(self.station_train_loaders[f"station_{station_number}_train_loaders"],
station_model=model, model=model,
global_model=self.globel_model_second[f"station_{station_number}_globalsecond"],
n_epochs=new_epoch[station_number], learning_rate=lr,
scheduler_milestones=s_m, scheduler_gamma=s_g,
LOG_FILE=self.log_files[station_number], emb_diff_to_server=self.emb_diff_to_server,
local_model=self.local_model_pth[station_number], round=rn + 1,
embedding_diff=self.embedding_diff,
embedding_diff_plus_loss=self.embedding_diff_plus_loss,
LOG_FILE_emb=self.log_emb_diff[station_number], station_n=station_number,
ls_function=self.lsfunction)
start_emb_df_accumulated[station_number] = emb_diff_epoch
results = self.get_parameters(model), len(self.station_train_loaders), {
"embedding_diff": start_emb_df_accumulated[station_number]}
else:
embedding_loss = train(self.station_train_loaders[f"station_{station_number}_train_loaders"],
station_model=model, model=model,
global_model=self.globel_model_second[f"station_{station_number}_globalsecond"],
n_epochs=new_epoch[station_number], learning_rate=lr,
scheduler_milestones=s_m, scheduler_gamma=s_g,
LOG_FILE=self.log_files[station_number], emb_diff_to_server=self.emb_diff_to_server,
local_model=self.local_model_pth[station_number], round=rn + 1,
embedding_diff=self.embedding_diff,
embedding_diff_plus_loss=self.embedding_diff_plus_loss,
LOG_FILE_emb=self.log_emb_diff[station_number], station_n=station_number,
ls_function=self.lsfunction)
results = self.get_parameters(model), len(self.station_train_loaders), {}
results_fed.append(results)
weights_epoch_diff[station_number] = embedding_loss.item()
aggregate_weights = self.fit(results_fed, rn)
root_folder = self.method + self.run_name
root_folder = os.path.join("../Output", root_folder)
if aggregate_weights is not None:
try:
state_dict = dict()
for idx, param_tensor in enumerate(self.global_model.state_dict()):
state_dict[param_tensor] = torch.from_numpy(
aggregate_weights[idx].cpu().numpy()) # Convert tensor to NumPy array
print(
"im saving model under following folder -----------------------------------------------------------------------")
model_dir = os.path.join(root_folder, "model")
os.makedirs(model_dir, exist_ok=True)
model_file = os.path.join(model_dir, f"round--{rn + 1}_weights.h5")
self.global_model.load_state_dict(state_dict, strict=False)
torch.save(self.global_model.state_dict(), model_file)
except Exception as e:
print(f"Error saving weights for round {rn + 1}: {e}")
for station_number, model in enumerate(self.global_model_stations.values(), start=1):
self.set_parameters(model, aggregate_weights)
if __name__ == "__main__":
station_properties = {
"permutatiion": [[0, 1, 2, 5], [0, 1, 2, 5]],
"label_list": [[0, 1, 2, 5], [0, 1, 2, 5]],
"similarity_labels": [[0, 1, 2, 3, 4, 5, 7, 10], [0, 1, 2, 4, 5, 6, 7, 10], [0, 1, 2, 4, 5, 7, 8, 10],
[0, 1, 2, 4, 5, 7, 9, 10]],
"n_way": 4,
"n_shot": 2,
"n_query": 8,
"n_tasks_per_epoch": 10, # 250
"n_validation_tasks": 1, # 150
"n_workers": 0,
"train_test_spl": "./workspace/Jammer_detection/Datasets/Diff_env_daten/base_data/train_random_dep_new.txt",
"root_dir": "./workspace/Jammer_detection/Datasets/Diff_env_daten/base_data/data",
"norm_vals": "./workspace/Jammer_detection/Datasets/norm_vals.npy",
"global_model_path": "./workspace/Jammer_detection/Src/Linkhalle_federated/Src/outputs/2024-04-10/trained_0123/trained_0123/model_weights0epoch100.h5"
}
##########traniner settings############
##define model
model = models.resnet18(pretrained=False)
model.fc = nn.Identity()
network_ = PrototypicalNetworks(model)
init_weights_pth = "./workspace/Jammer_detection/Src/federated_synthetic_data/outputs/2024-02-27/18-12-02/model_global_synthetic/model_weights0epoch100.h5"
parser = argparse.ArgumentParser(description="Federated Training Argument Parser")
parser.add_argument("--method", type=str, required=True, help="Training method")
parser.add_argument("--run_name", type=str, required=True, help="Name of the run")
parser.add_argument("--model", type=str, default=network_, help="Model name")
parser.add_argument("--station_properties", default=station_properties, help="Properties for clients")
parser.add_argument("--rounds", type=int, default=10, help="Number of rounds")
parser.add_argument("--epochs", type=int, default=5, help="Number of epochs")
parser.add_argument("--num_clients", type=int, default=2, help="Number of clients")
parser.add_argument("--init_weights_pth", type=str, default=init_weights_pth, help="initial weights path")
parser.add_argument("--weighted_loss", type=bool, default=False, help="weighted_loss")
parser.add_argument("--embedding_diff", type=bool, default=False, help="embedding_diff")
parser.add_argument("--embedding_diff_plus_loss", type=bool, default=False, help="embedding_diff_plus_loss")
parser.add_argument("--emb_diff_to_server", type=bool, default=False, help="emb_diff_to_server")
parser.add_argument("--momentum_vector", type=int, default=None, help="momentum vector")
parser.add_argument("--lossfunction", type=str, default=None, help="loss function")
parser.add_argument("--env", type=list, default=[[2, 3, 11, 4, 5], [6, 7, 8, 9, 10]], help="env list")
parser.add_argument("--mode", type=bool, default=False, help="epoch difference")
# Add more arguments as needed
args = parser.parse_args()
print(f"embeding diff to server value is : {args.emb_diff_to_server}")
federated_training = federated_training(
args.method,
args.run_name,
args.model,
args.station_properties,
args.rounds,
args.epochs,
args.num_clients,
args.init_weights_pth,
args.weighted_loss,
args.embedding_diff,
args.embedding_diff_plus_loss,
args.emb_diff_to_server,
args.momentum_vector,
args.lossfunction,
args.env,
args.mode
)
federated_training.train_fedretaed()
|