opt = arg_parser.parse_args()
arg_parser.create_log_path(opt, add_path_var="linear_model")
opt.SVM_training_samples = 20
# Device configuration
opt.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# random seeds
torch.manual_seed(opt.seed)
torch.cuda.manual_seed(opt.seed)
np.random.seed(opt.seed)
# load pretrained model
context_model, _ = load_audio_model.load_model_and_optimizer(
opt, reload_model=True
)
context_model.eval()
_, train_dataset, _, test_dataset = get_dataloader.get_libri_dataloaders(opt)
try:
# Train the model
speaker_svm, speaker_id_dict, training_speaker_keys = SVM_train(
opt, context_model, train_dataset
)
# Test the model
accuracy = SVM_test(
opt,
speaker_id_dict,
logs.create_log(model, epoch=epoch, optimizer=optimizer)
if __name__ == "__main__":
opt = arg_parser.parse_args()
arg_parser.create_log_path(opt)
# set random seeds
torch.manual_seed(opt.seed)
torch.cuda.manual_seed(opt.seed)
np.random.seed(opt.seed)
random.seed(opt.seed)
# load model
model, optimizer = load_audio_model.load_model_and_optimizer(opt)
# initialize logger
logs = logger.Logger(opt)
# get datasets and dataloaders
train_loader, train_dataset, test_loader, test_dataset = get_dataloader.get_libri_dataloaders(
opt
)
try:
# Train the model
train(opt, model)
except KeyboardInterrupt:
print("Training got interrupted, saving log-files now.")
opt.num_epochs = 50
opt.learning_rate = 1e-3
arg_parser.create_log_path(opt, add_path_var="linear_model")
# random seeds
torch.manual_seed(opt.seed)
torch.cuda.manual_seed(opt.seed)
np.random.seed(opt.seed)
## load model
context_model, optimizer = load_audio_model.load_model_and_optimizer(
opt,
reload_model=True,
calc_accuracy=True,
num_GPU=1,
)
context_model.eval()
n_features = context_model.module.reg_hidden
loss = loss_supervised_speaker.Speaker_Loss(
opt, n_features, calc_accuracy=True
)
optimizer = torch.optim.Adam(loss.parameters(), lr=opt.learning_rate)
# load dataset
train_loader, _, test_loader, _ = get_dataloader.get_libri_dataloaders(opt)