from common_dl import set_random_seeds from common_dsp import * from gesture.models.d2l_resnet import d2lresnet from myskorch import on_epoch_begin_callback, on_batch_end_callback from ecog_finger.config import * from ecog_finger.preprocess.chn_settings import get_channel_setting os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2' try: mne.set_config('MNE_LOGGING_LEVEL', 'ERROR') except TypeError as err: print(err) seed = 20200220 # random seed to make results reproducible set_random_seeds(seed=seed) import inspect as i import sys #sys.stdout.write(i.getsource(deepnet)) sid = 2 fs = 1000 use_active_only = False if use_active_only: active_chn = get_channel_setting(sid) else: active_chn = 'all' filename = data_dir + 'fingerflex/data/' + str(sid) + '/' + str( sid) + '_fingerflex.mat'
shuffle=True, pin_memory=False) val_loader = DataLoader(dataset=val_ds, batch_size=batch_size, pin_memory=False) #test it #(x,y)=iter(train_loader).next() # x: torch.Size([1, 10, 148, 250]); y: torch.Size([1]) X_test.shape len(val_loader) * 4 cuda = torch.cuda.is_available() device = 'cuda' if cuda else 'cpu' seed = 20200220 set_random_seeds(seed=seed) # same as braindecode random seeding net = my_resnet18(chn_num, 5, pretrained=False, logsoftmax=False).float() if cuda: net.cuda() #x= torch.randn(1, 10, 148, 250) #net(x).shape lr = 0.0001 weight_decay = 1e-10 batch_size = 4 epoch_num = 500 #criterion=torch.nn.NLLLoss() #optimizer = torch.optim.Adam(net.parameters(), lr=lr)