def __init__(self, input_size, input_channels, feature_dim, num_classes, use_l2_norm, use_dropout, use_batch_norm, device='cuda'): super(SimpleClassifier, self).__init__() self.input_size = input_size self.num_classes = num_classes self.use_l2_norm = use_l2_norm self.use_dropout = use_dropout self.use_batch_norm = use_batch_norm self.input_channels = input_channels self.feature_dim = feature_dim self.device = device self.encoder = Encoder(input_size, input_channels, feature_dim) final_fc = [] if use_batch_norm: final_fc.append(nn.BatchNorm1d(feature_dim)) if use_dropout: final_fc.append(nn.Dropout(0.5)) final_fc.append(nn.Linear(feature_dim, num_classes)) self.final_fc = nn.Sequential(*final_fc)
def __init__(self, input_size, input_channels, feature_dim, pred_steps, num_class, use_l2_norm, use_dropout, use_batch_norm, device): super(CPCClassifier, self).__init__() self.input_size = input_size self.input_channels = input_channels self.feature_dim = feature_dim self.pred_steps = pred_steps self.device = device self.use_l2_norm = use_l2_norm self.use_dropout = use_dropout self.use_batch_norm = use_batch_norm self.encoder = Encoder(input_size, input_channels, feature_dim) # self.agg = nn.GRU(input_size=feature_dim, hidden_size=feature_dim, batch_first=True) # self.relu = nn.ReLU(inplace=True) final_fc = [] if use_batch_norm: final_fc.append(nn.BatchNorm1d(self.feature_size)) if use_dropout: final_fc.append(nn.Dropout(0.5)) final_fc.append(nn.Linear(feature_dim, num_class)) self.final_fc = nn.Sequential(*final_fc)
def __init__(self, input_size, input_channels, hidden_channels, feature_dim, device='cuda'): super(TemporalShuffling, self).__init__() self.input_size = input_size self.input_channels = input_channels self.hidden_channels = hidden_channels self.feature_dim = feature_dim self.device = device self.encoder = Encoder(input_size, input_channels, feature_dim) self.linear_head = nn.Linear(2 * feature_dim, 1, bias=True)
def __init__(self, input_size, input_channels, hidden_channels, feature_dim, device='cuda'): super(RelativePosition, self).__init__() self.input_size = input_size self.input_channels = input_channels self.hidden_channels = hidden_channels self.feature_dim = feature_dim self.device = device self.encoder = Encoder(input_size, input_channels, feature_dim) self.linear_head = nn.Linear(feature_dim, 1, bias=True)
def run(run_id, train_patients, test_patients, args): print('Train patient ids:', train_patients) print('Test patient ids:', test_patients) if args.data_name == 'SEED': input_size = 200 elif args.data_name == 'DEAP': input_size = 128 elif args.data_name == 'AMIGOS': input_size = 128 else: raise ValueError encoder = Encoder(input_size=input_size, input_channel=args.input_channel, feature_dim=args.feature_dim) encoder.cuda(args.device) discriminator = Discriminator(args.feature_dim, args.device) discriminator.cuda(args.device) dataset = TNCDataset() pretrain(run_id, encoder, discriminator, dataset, args.device, args)
def __init__(self, input_size, input_channels, hidden_channels, feature_dim, device='cuda'): super(TLoss, self).__init__() self.input_size = input_size self.input_channels = input_channels self.hidden_channels = hidden_channels self.feature_dim = feature_dim self.device = device self.encoder = Encoder(input_size, input_channels, feature_dim)
def __init__(self, input_size, input_channels, feature_dim, pred_steps, use_temperature, temperature, device): super(DPC, self).__init__() self.input_size = input_size self.input_channels = input_channels self.feature_dim = feature_dim self.pred_steps = pred_steps self.use_temperature = use_temperature self.temperature = temperature self.device = device self.encoder = Encoder(input_size, input_channels, feature_dim) self.agg = nn.GRU(input_size=feature_dim, hidden_size=feature_dim, batch_first=True) self.predictor = nn.Linear(feature_dim, feature_dim) self.relu = nn.ReLU(inplace=True) self.targets = None