client_lr = 0.0316
server_opt = "Yogi"
client_opt = "SGD"
client_opt_strategy = "reinit"
# image_norm = "tflike"
# TODO a paraméterek helytelen nevére nem adott hibát
config = TorchFederatedLearnerCIFAR100Config(
BREAK_ROUND=1500,
CLIENT_LEARNING_RATE=client_lr,
CLIENT_OPT=client_opt,
# CLIENT_OPT_ARGS=common.get_args(client_opt),
CLIENT_OPT_L2=1e-4,
CLIENT_OPT_STRATEGY=client_opt_strategy,
SERVER_OPT=server_opt,
SERVER_OPT_ARGS=common.get_args(server_opt),
SERVER_LEARNING_RATE=server_lr,
IS_IID_DATA=is_iid,
BATCH_SIZE=B,
CLIENT_FRACTION=C,
N_CLIENTS=NC,
N_EPOCH_PER_CLIENT=E,
MAX_ROUNDS=max_rounds,
IMAGE_NORM="recordwisefull",
NORM="group",
INIT="tffed",
AUG="basicf")
config_technical = TorchFederatedLearnerTechnicalConfig(BREAK_ROUND=300,
EVAL_ROUND=100)
name = f"{config.SERVER_OPT}: {config.SERVER_LEARNING_RATE} - {config.CLIENT_OPT_STRATEGY} - {config.CLIENT_OPT}: {config.CLIENT_LEARNING_RATE}"
experiment = Experiment(workspace="federated-learning",
project_name=project_name)
common.do_training(experiment, name, config, config_technical)
# TODO a paraméterek helytelen nevére nem adott hibát
for l2 in [1e-3, 1e-2, 1e-1, 1e-1, 1e1]:
s_opt_args = common.get_args(server_opt)
s_opt_args["weight_decay"] = l2
config = TorchFederatedLearnerCIFAR100Config(
BREAK_ROUND=300,
CLIENT_LEARNING_RATE=client_lr,
CLIENT_OPT=client_opt,
CLIENT_OPT_ARGS=common.get_args(client_opt),
CLIENT_OPT_L2=l2,
CLIENT_OPT_STRATEGY=client_opt_strategy,
SERVER_OPT=server_opt,
SERVER_OPT_ARGS=s_opt_args,
SERVER_LEARNING_RATE=server_lr,
IS_IID_DATA=is_iid,
BATCH_SIZE=B,
CLIENT_FRACTION=C,
N_CLIENTS=NC,
N_EPOCH_PER_CLIENT=E,
MAX_ROUNDS=max_rounds,
DL_N_WORKER=0,
NORM="group",
# IMAGE_NORM=image_norm,
INIT="tffed",
)
config_technical = TorchFederatedLearnerTechnicalConfig(HIST_SAMPLE=0)
name = f"{config.SERVER_OPT}: {config.SERVER_LEARNING_RATE} - {config.CLIENT_OPT_STRATEGY} - {config.CLIENT_OPT}: {config.CLIENT_LEARNING_RATE}"
experiment = Experiment(workspace="federated-learning",
project_name=project_name)
common.do_training(experiment, name, config, config_technical)
]
config_changes = [
("SGD", "SGD", 1, 0.1, "nothing"),
("Yogi", "SGD", 0.1, 0.01, "nothing"),
("Yogi", "Yogi", 0.1, 0.0001, "avg"),
("Yogi", "Yogi", 0.1, 0.0001, "reinit"),
("Yogi", "Yogi", 0.1, 0.0001, "nothing"),
]
for values in config_changes:
config = TorchFederatedLearnerCIFAR100Config(
BREAK_ROUND=1500,
CLIENT_OPT_L2=1e-4,
IS_IID_DATA=is_iid,
BATCH_SIZE=B,
CLIENT_FRACTION=C,
N_CLIENTS=NC,
N_EPOCH_PER_CLIENT=E,
MAX_ROUNDS=max_rounds,
IMAGE_NORM="recordwisefull",
NORM="group",
INIT="tffed",
AUG="flipf",
)
for k, v in zip(param_names, values):
setattr(config, k, v)
config_technical = TorchFederatedLearnerTechnicalConfig(
SAVE_CHP_INTERVALL=5)
name = f"{config.SERVER_OPT}: {config.SERVER_LEARNING_RATE} - {config.CLIENT_OPT_STRATEGY} - {config.CLIENT_OPT}: {config.CLIENT_LEARNING_RATE}"
experiment = Experiment(workspace="federated-learning",
project_name=project_name)
common.do_training(experiment, name, config, config_technical)
# image_norm = "tflike"
# TODO a paraméterek helytelen nevére nem adott hibát
for init in ["tffed"]: #TorchInitRepo.get_opt_names():
config = TorchFederatedLearnerCIFAR100Config(
BREAK_ROUND=300,
CLIENT_LEARNING_RATE=client_lr,
CLIENT_OPT=client_opt,
# CLIENT_OPT_ARGS=common.get_args(client_opt),
# CLIENT_OPT_L2=1e-4,
CLIENT_OPT_STRATEGY=client_opt_strategy,
SERVER_OPT=server_opt,
SERVER_OPT_ARGS={
"betas": (0.9, 0.999),
"initial_accumulator": 0.0
},
SERVER_LEARNING_RATE=server_lr,
IS_IID_DATA=is_iid,
BATCH_SIZE=B,
CLIENT_FRACTION=C,
N_CLIENTS=NC,
N_EPOCH_PER_CLIENT=E,
MAX_ROUNDS=max_rounds,
DL_N_WORKER=0,
NORM="group",
# IMAGE_NORM=image_norm,
INIT=init,
)
config_technical = TorchFederatedLearnerTechnicalConfig(HIST_SAMPLE=0)
name = f"{config.SERVER_OPT}: {config.SERVER_LEARNING_RATE} - {config.CLIENT_OPT_STRATEGY} - {config.CLIENT_OPT}: {config.CLIENT_LEARNING_RATE}"
experiment = Experiment(workspace="federated-learning",
project_name=project_name)
