def train(net: EMGClassifier, gesture_amount):
train_set = CSLDataset(gesture=gesture_amount,
frame_x=True,
sequence_len=10,
train=True)
net.dataset = train_set
net.fit_with_dataset()
return net
def train(net: EMGClassifier, gesture_indices: list):
train_set = CapgDataset(gestures_label_map=net.gesture_map,
sequence_len=1,
frame_x=True,
gesture_list=gesture_indices,
train=True)
net.dataset = train_set
net.fit_with_dataset()
return net
def train(net: EMGClassifier, gesture_indices: list):
if net.hyperparamters['dataset'] == 'capg':
train_set = CapgDataset(gestures_label_map=net.gesture_map,
sequence_len=1,
gesture_list=gesture_indices,
train=True)
else:
train_set = CSLDataset(gesture=8,
sequence_len=1,
train=True)
net.dataset = train_set
net.fit_with_dataset()
return net
def main(train_args, TEST_MODE=False):
args = train_args
if args['gesture_num'] == 8:
all_gestures = list(range(8))
elif args['gesture_num'] == 12:
all_gestures = list(range(8, 20))
else:
all_gestures = list(range(args['gesture_num']))
model = CNN(len(all_gestures))
name = args['name']
sub_folder = args['sub_folder']
tensorboard_cb = config_tensorboard(name, sub_folder)
from emg.utils.lr_scheduler import DecayLR
lr_callback = DecayLR(start_lr=args['lr'],
gamma=0.5,
step_size=args['lr_step'])
net = EMGClassifier(module=model,
model_name=name,
sub_folder=sub_folder,
hyperparamters=args,
optimizer=torch.optim.Adam,
gesture_list=all_gestures,
callbacks=[tensorboard_cb, lr_callback])
if not TEST_MODE:
net = train(net, all_gestures)
confusion_matrx = test(net, all_gestures)
return confusion_matrx
def main(train_args, TEST_MODE=False):
# 1. 设置好optimizer
# 2. 定义好model
args = train_args
all_gestures = list(range(args['gesture_num']))
model = C3D(len(all_gestures))
name = args['name']
sub_folder = args['sub_folder']
# from emg.utils import config_tensorboard
# tensorboard_cb = config_tensorboard(name, sub_folder)
from emg.utils.lr_scheduler import DecayLR
lr_callback = DecayLR(start_lr=args['lr'],
gamma=0.5,
step_size=args['lr_step'])
net = EMGClassifier(module=model,
model_name=name,
sub_folder=sub_folder,
hyperparamters=args,
optimizer=torch.optim.Adam,
gesture_list=all_gestures,
callbacks=[lr_callback])
net = train(net, args['gesture_num'])
net = test(net, args['gesture_num'], all_gestures)
def test(net: EMGClassifier, gesture_amount, gesture_indices: list):
test_set = CSLDataset(gesture=gesture_amount,
frame_x=True,
sequence_len=10,
train=False)
avg_score = net.test_model(gesture_indices, test_set)
print('test accuracy: {:.4f}'.format(avg_score))
return net
def test(net: EMGClassifier, gesture_indices: list):
test_set = CapgDataset(gestures_label_map=net.gesture_map,
sequence_len=1,
frame_x=True,
gesture_list=gesture_indices,
train=False)
avg_score, matrix = net.test_model(gesture_indices, test_set)
print('test accuracy: {:.4f}'.format(avg_score))
return matrix
def test(net: EMGClassifier, gesture_indices: list):
if net.hyperparamters['dataset'] == 'capg':
test_set = CapgDataset(gestures_label_map=net.gesture_map,
sequence_len=1,
gesture_list=gesture_indices,
train=False)
else:
test_set = CSLDataset(gesture=8,
sequence_len=1,
train=False)
avg_score, matrix = net.test_model(gesture_indices, test_set)
print('test accuracy: {:.4f}'.format(avg_score))
return matrix
def main(train_args, TEST_MODE=False):
if train_args['dataset'] == 'capg':
args = {**train_args, **capg_args}
else:
args = {**train_args, **csl_args}
if args['gesture_num'] == 8:
all_gestures = list(range(8))
elif args['gesture_num'] == 12:
all_gestures = list(range(8, 20))
else:
all_gestures = list(range(args['gesture_num']))
model = MLP(args['input_size'], args['hidden_size'], len(all_gestures))
name = args['name']
sub_folder = args['sub_folder']
# from emg.utils import config_tensorboard
# tensorboard_cb = config_tensorboard(name, sub_folder, model, (1, 128))
from emg.utils.lr_scheduler import DecayLR
lr_callback = DecayLR(start_lr=args['lr'], gamma=0.5, step_size=args['lr_step'])
net = EMGClassifier(module=model,
model_name=name,
sub_folder=sub_folder,
hyperparamters=args,
optimizer=torch.optim.Adam,
gesture_list=all_gestures,
callbacks=[lr_callback])
if not TEST_MODE:
net = train(net, all_gestures)
confusion_matrx = test(net, all_gestures)
return confusion_matrx