def __init__(self, args):
self.run_name = args.run_name + '_' + str(time.time()).split('.')[0]
self.current_run_basepath = args.network_metrics_basepath + '/' + self.run_name + '/'
self.learning_rate = args.learning_rate
self.epochs = args.epochs
self.test_net = args.test_net
self.train_net = args.train_net
self.batch_size = args.batch_size
self.num_classes = args.num_classes
self.audio_basepath = args.audio_basepath
self.train_data_file = args.train_data_file
self.test_data_file = args.test_data_file
self.data_read_path = args.data_save_path
self.is_cuda_available = torch.cuda.is_available()
self.display_interval = args.display_interval
self.sampling_rate = args.sampling_rate
self.sample_size_in_seconds = args.sample_size_in_seconds
self.overlap = args.overlap
self.network_metrics_basepath = args.network_metrics_basepath
self.tensorboard_summary_path = self.current_run_basepath + args.tensorboard_summary_path
self.network_save_path = self.current_run_basepath + args.network_save_path
self.network_restore_path = args.network_restore_path
self.device = torch.device("cuda" if self.is_cuda_available else "cpu")
self.network_save_interval = args.network_save_interval
self.normalise = args.normalise_while_training
self.data_augment = args.data_augment
self.dropout = args.dropout
self.threshold = args.threshold
self.debug_filename = self.current_run_basepath + '/' + args.debug_filename
paths = [self.network_save_path, self.tensorboard_summary_path]
file_utils.create_dirs(paths)
self.pos_weight = None
self.learning_rate_decay = args.learning_rate_decay
self._min, self._max = float('inf'), -float('inf')
self.logger = Logger(name=self.run_name,
log_path=self.network_save_path).get_logger()
self.logger.info(str(json.dumps(args, indent=4)))
self.writer = SummaryWriter(self.tensorboard_summary_path)
self.batch_loss, self.batch_accuracy, self.uar = [], [], []
self.logger.info(f'Configs used:\n{json.dumps(args, indent=4)}')
self.data_loaders = {}
self.dataset_sizes = {}
def __init__(self, args):
self.run_name = args.run_name + '_' + str(time.time()).split('.')[0]
self.current_run_basepath = args.network_metrics_basepath + '/' + self.run_name + '/'
self.learning_rate = args.learning_rate
self.epochs = args.epochs
self.test_net = args.test_net
self.train_net = args.train_net
self.batch_size = args.batch_size
self.num_classes = args.num_classes
self.audio_basepath = args.audio_basepath
self.train_data_file = args.train_data_file
self.test_data_file = args.test_data_file
self.data_read_path = args.data_save_path
self.is_cuda_available = torch.cuda.is_available()
self.display_interval = args.display_interval
self.sampling_rate = args.sampling_rate
self.sample_size_in_seconds = args.sample_size_in_seconds
self.overlap = args.overlap
self.alpha = args.alpha
self.beta = args.beta
self.network_metrics_basepath = args.network_metrics_basepath
self.tensorboard_summary_path = self.current_run_basepath + args.tensorboard_summary_path
self.network_save_path = self.current_run_basepath + args.network_save_path
self.network_restore_path = args.network_restore_path
self.device = torch.device("cuda" if self.is_cuda_available else "cpu")
self.network_save_interval = args.network_save_interval
self.normalise = args.normalise_while_training
self.dropout = args.dropout
self.threshold = args.threshold
paths = [self.network_save_path, self.tensorboard_summary_path]
file_utils.create_dirs(paths)
self.network = ConvAutoEncoder().to(self.device)
self.reconstruction_loss = nn.BCEWithLogitsLoss()
self.optimiser = optim.Adam(self.network.parameters(), lr=self.learning_rate)
self._min, self._max = float('inf'), -float('inf')
if self.train_net:
self.network.train()
self.log_file = open(self.network_save_path + '/' + self.run_name + '.log', 'w')
self.log_file.write(json.dumps(args))
if self.test_net:
print('Loading Network')
self.network.load_state_dict(torch.load(self.network_restore_path, map_location=self.device))
self.network.eval()
self.log_file = open(self.network_restore_path.replace('_40.pt', '.log'), 'a')
print('\n\n\n********************************************************', file=self.log_file)
print('Testing Model - ', self.network_restore_path)
print('Testing Model - ', self.network_restore_path, file=self.log_file)
print('********************************************************', file=self.log_file)
self.writer = SummaryWriter(self.tensorboard_summary_path)
print("Network config:\n", self.network)
print("Network config:\n", self.network, file=self.log_file)
self.batch_accuracy, self.uar = [], []
print('Configs used:\n', json.dumps(args, indent=4))
print('Configs used:\n', json.dumps(args, indent=4), file=self.log_file)
def __init__(self, args):
self.run_name = args.run_name + '_' + str(time.time()).split('.')[0]
self.current_run_basepath = args.network_metrics_basepath + '/' + self.run_name + '/'
self.learning_rate = args.learning_rate
self.epochs = args.epochs
self.test_net = args.test_net
self.train_net = args.train_net
self.batch_size = args.batch_size
self.num_classes = args.num_classes
self.audio_basepath = args.audio_basepath
self.train_data_file = args.train_data_file
self.test_data_file = args.test_data_file
self.data_read_path = args.data_save_path
self.is_cuda_available = torch.cuda.is_available()
self.display_interval = args.display_interval
self.sampling_rate = args.sampling_rate
self.sample_size_in_seconds = args.sample_size_in_seconds
self.overlap = args.overlap
self.network_metrics_basepath = args.network_metrics_basepath
self.tensorboard_summary_path = self.current_run_basepath + args.tensorboard_summary_path
self.network_save_path = self.current_run_basepath + args.network_save_path
self.network_restore_path = args.network_restore_path
self.device = torch.device("cuda" if self.is_cuda_available else "cpu")
self.network_save_interval = args.network_save_interval
self.normalise = args.normalise_while_training
self.dropout = args.dropout
self.threshold = args.threshold
self.debug_filename = self.current_run_basepath + '/' + args.debug_filename
paths = [self.network_save_path, self.tensorboard_summary_path]
file_utils.create_dirs(paths)
args.sincnet_saved_model = args.data_save_path + args.sincnet_saved_model
self.sincnet_params = []
self.network = SincNet(args).to(self.device)
# for x in self.network.state_dict().keys():
# # if isinstance(self.network.state_dict()[x], float):
# print(x, torch.max(self.network.state_dict()[x]), torch.min(self.network.state_dict()[x]))
self.saved_model = \
torch.load(args['sincnet_saved_model'], map_location=self.device)[
'CNN_model_par']
self.load_my_state_dict(self.saved_model)
print(self.sincnet_params)
# exit()
# for i, param in enumerate(self.network.parameters()):
# if name in self.sincnet_params:
# self.network.state_dict()[name].requires_grad = False
# if i <= 19:
# param.requires_grad = False
# for x in self.network.state_dict().keys():
# # if isinstance(self.network.state_dict()[x], float):
# print(x, torch.max(self.network.state_dict()[x]), torch.min(self.network.state_dict()[x]))
self.pos_weight = None
self.loss_function = None
self.learning_rate_decay = args.learning_rate_decay
self.optimiser = optim.RMSprop(self.network.parameters(),
lr=self.learning_rate,
alpha=0.95,
eps=1e-8)
self.scheduler = torch.optim.lr_scheduler.ExponentialLR(
self.optimiser, gamma=self.learning_rate_decay)
self._min, self._max = float('inf'), -float('inf')
if self.train_net:
self.network.train()
self.log_file = open(
self.network_save_path + '/' + self.run_name + '.log', 'w')
self.log_file.write(json.dumps(args))
if self.test_net:
print('Loading Network')
self.network.load_state_dict(
torch.load(self.network_restore_path,
map_location=self.device))
self.network.eval()
self.log_file = open(
self.network_restore_path.replace('_40.pt', '.log'), 'a')
print(
'\n\n\n********************************************************',
file=self.log_file)
print('Testing Model - ', self.network_restore_path)
print('Testing Model - ',
self.network_restore_path,
file=self.log_file)
print('********************************************************',
file=self.log_file)
self.writer = SummaryWriter(self.tensorboard_summary_path)
print("Network config:\n", self.network)
print("Network config:\n", self.network, file=self.log_file)
self.batch_loss, self.batch_accuracy, self.uar = [], [], []
print('Configs used:\n', json.dumps(args, indent=4))
print('Configs used:\n',
json.dumps(args, indent=4),
file=self.log_file)
def __init__(self, args):
self.run_name = args.run_name + '_' + str(time.time()).split('.')[0]
self.current_run_basepath = args.network_metrics_basepath + '/' + self.run_name + '/'
self.learning_rate = args.learning_rate
self.epochs = args.epochs
self.test_net = args.test_net
self.train_net = args.train_net
self.batch_size = args.batch_size
self.num_classes = args.num_classes
self.audio_basepath = args.audio_basepath
self.train_data_file = args.train_data_file
self.test_data_file = args.test_data_file
self.data_read_path = args.data_save_path
self.is_cuda_available = torch.cuda.is_available()
self.display_interval = args.display_interval
self.sampling_rate = args.sampling_rate
self.sample_size_in_seconds = args.sample_size_in_seconds
self.overlap = args.overlap
self.network_metrics_basepath = args.network_metrics_basepath
self.tensorboard_summary_path = self.current_run_basepath + args.tensorboard_summary_path
self.network_save_path = self.current_run_basepath + args.network_save_path
self.network_restore_path = args.network_restore_path
self.device = torch.device("cuda" if self.is_cuda_available else "cpu")
self.network_save_interval = args.network_save_interval
self.normalise = args.normalise_while_training
self.data_augment = args.data_augment
self.dropout = args.dropout
self.threshold = args.threshold
self.debug_filename = self.current_run_basepath + '/' + args.debug_filename
paths = [self.network_save_path, self.tensorboard_summary_path]
file_utils.create_dirs(paths)
self.network = ConvNet().to(self.device)
self.pos_weight = None
self.loss_function = None
self.learning_rate_decay = args.learning_rate_decay
self.optimiser = optim.AdamW(self.network.parameters(),
lr=self.learning_rate)
self.scheduler = torch.optim.lr_scheduler.ExponentialLR(
self.optimiser, gamma=self.learning_rate_decay)
self._min, self._max = float('inf'), -float('inf')
if self.train_net:
self.network.train()
self.logger = Logger(name=self.run_name,
log_path=self.network_save_path).get_logger()
self.logger.info(str(json.dumps(args, indent=4)))
if self.test_net:
self.logger.info('Loading Network')
self.network.load_state_dict(
torch.load(self.network_restore_path,
map_location=self.device))
self.network.eval()
self.logger.info(
'\n\n\n********************************************************'
)
self.logger.info(f'Testing Model - {self.network_restore_path}')
self.logger.info(
'********************************************************')
self.writer = SummaryWriter(self.tensorboard_summary_path)
self.logger.info(f"Network Architecture:\n,{self.network}")
self.batch_loss, self.batch_accuracy, self.uar = [], [], []
self.logger.info(f'Configs used:\n{json.dumps(args, indent=4)}')