def main():
"""
Main function to spawn the train and test process.
"""
args = parse_args()
cfg = load_config(args)
print('=' * 20)
# print(cfg)
print('Num of GPUs: ', cfg.NUM_GPUS)
print(cfg.TRAIN)
print(cfg.TEST)
print('output dir is: ', cfg.OUTPUT_DIR)
# Perform training.
if cfg.TRAIN.ENABLE:
print("begin to trian the model... ")
if cfg.NUM_GPUS > 1:
print('gpu is over 1')
torch.multiprocessing.spawn(
mpu.run,
nprocs=cfg.NUM_GPUS,
args=(
cfg.NUM_GPUS,
train,
args.init_method,
cfg.SHARD_ID,
cfg.NUM_SHARDS,
cfg.DIST_BACKEND,
cfg,
),
daemon=False,
)
else:
train(cfg=cfg)
# Perform multi-clip testing.
if cfg.TEST.ENABLE:
print("begin to test the model... ")
if cfg.NUM_GPUS > 1:
torch.multiprocessing.spawn(
mpu.run,
nprocs=cfg.NUM_GPUS,
args=(
cfg.NUM_GPUS,
test,
args.init_method,
cfg.SHARD_ID,
cfg.NUM_SHARDS,
cfg.DIST_BACKEND,
cfg,
),
daemon=False,
)
else:
test(cfg=cfg)
def main():
"""
Main function to spawn the train and test process.
"""
args = parse_args()
cfg = load_config(args)
# Perform training.
print("Number of GPUS: ", cfg.NUM_GPUS)
if cfg.TRAIN.ENABLE:
if cfg.NUM_GPUS > 1:
torch.multiprocessing.spawn(
mpu.run,
nprocs=cfg.NUM_GPUS,
args=(
cfg.NUM_GPUS,
train,
args.init_method,
cfg.SHARD_ID,
cfg.NUM_SHARDS,
cfg.DIST_BACKEND,
cfg,
),
daemon=False,
)
else:
train(cfg=cfg)
# Perform multi-clip testing.
if cfg.TEST.ENABLE:
if cfg.NUM_GPUS > 1:
torch.multiprocessing.spawn(
mpu.run,
nprocs=cfg.NUM_GPUS,
args=(
cfg.NUM_GPUS,
test,
args.init_method,
cfg.SHARD_ID,
cfg.NUM_SHARDS,
cfg.DIST_BACKEND,
cfg,
),
daemon=False,
)
else:
test(cfg=cfg)
# squared hinge loss
loss = T.mean(T.sqr(T.maximum(0., 1. - target * train_output)))
params = lasagne.layers.get_all_params(cnn, trainable=True)
updates = lasagne.updates.adam(loss_or_grads=loss, params=params, learning_rate=LR)
test_output = lasagne.layers.get_output(cnn, deterministic=True)
test_loss = T.mean(T.sqr(T.maximum(0., 1. - target * test_output)))
test_err = T.mean(T.neq(T.argmax(test_output, axis=1), T.argmax(target, axis=1)), dtype=theano.config.floatX)
# Compile a function performing a training step on a mini-batch (by giving the updates dictionary)
# and returning the corresponding training loss:
train_fn = theano.function([input, target, LR], loss, updates=updates)
# Compile a second function computing the validation loss and accuracy:
val_fn = theano.function([input, target], [test_loss, test_err])
print('Training...')
train_net.train(
train_fn, val_fn,
cnn,
batch_size,
LR_start, LR_decay,
num_epochs,
train_set.X, train_set.y,
valid_set.X, valid_set.y,
test_set.X, test_set.y,
shuffle_parts=shuffle_parts)
# load your own csv or other format data here
train_file = os.path.join(current_path, 'xxx.csv')
# load the feature npy to the memory if possible(for a faster feature reading speed)
feature_npy = 'xxx.npy'
# data loading
train_pair = data_process.get_data_pair(train_file)
# pre-loding the feature to the memory
c3d_feature = data_process.load_feature(feature_npy)
pair = {'train': train_pair}
show_datasets = {x: c3d_datasets(pair[x], c3d_feature, x) for x in ['train']}
show_data_sizes = {x: len(show_datasets[x]) for x in ['train']}
dataloaders = {x: DataLoader(show_datasets[x], batch_size = 256, shuffle = True, num_workers = 4)
for x in ['train']}
use_gpu = torch.cuda.is_available()
# initilize the network and the optimizer stretigies
model = embedding_net()
net = triplet_net(model)
net = net.cuda()
criterion = torch.nn.MarginRankingLoss(margin = 0.5)
optimizer = optim.SGD(net.parameters(), lr = 0.01, momentum = 0.9)
exp_lr_scheduler = lr_scheduler.StepLR(optimizer, step_size = 10, gamma = 0.1)
train(dataloaders, net, criterion, optimizer, exp_lr_scheduler, show_data_sizes, use_gpu, num_epoches = 100)
def main():
# 设定训练网络所用的硬件设备
device = torch.device('cuda')
# 选择网络模型
# net = EEGNet1_3().to(device)
net = EEGNet2018().to(device)
# init all
# 初始化总轮回数，用来显示训练曲线
global_step = 0
# 初始化网络权重
net = train_net.weights_init(net)
# 是否可视化，1 or 0
vis = 1
if vis == 1:
viz = Visdom()
if vis == 1:
viz.line([0.], [0.], win='train_loss', update='append', opts={'title': 'train_loss'})
# viz.line([0.], [0.], win='test_results', update='append', opts={'title': 'test_results'})
# %%
"""loaddata"""
dataset_A_train = {}
targets_A_train = {}
dataset_A_hold = {}
targets_A_hold = {}
filename_train = "G:\EEGNet\data\EEGDataset\ERP\TrainDatas/Circle/dataset_A_train.mat"
filename_label = "G:\EEGNet\data\EEGDataset\ERP\TrainDatas/Circle/dataset_A_label.mat"
filename_test = "G:\EEGNet\data\EEGDataset\ERP\TestDatas/Circle/dataset_A_test.mat"
filename_test_label = "G:\EEGNet\data\EEGDataset\ERP\TestDatas/Circle/dataset_A_label.mat"
# %%
dataset_A_train[0] = scipy.io.loadmat(filename_train)['Tdata1']
targets_A_train[0] = scipy.io.loadmat(filename_label)['Tlabel1']
dataset_A_train[1] = scipy.io.loadmat(filename_train)['Tdata2']
targets_A_train[1] = scipy.io.loadmat(filename_label)['Tlabel2']
dataset_A_train[2] = scipy.io.loadmat(filename_train)['Tdata3']
targets_A_train[2] = scipy.io.loadmat(filename_label)['Tlabel3']
dataset_A_train[3] = scipy.io.loadmat(filename_train)['Tdata4']
targets_A_train[3] = scipy.io.loadmat(filename_label)['Tlabel4']
dataset_A_train[4] = scipy.io.loadmat(filename_train)['Tdata5']
targets_A_train[4] = scipy.io.loadmat(filename_label)['Tlabel5']
dataset_A_hold[0] = scipy.io.loadmat(filename_train)['Vdata1']
targets_A_hold[0] = scipy.io.loadmat(filename_label)['Vlabel1']
dataset_A_hold[1] = scipy.io.loadmat(filename_train)['Vdata2']
targets_A_hold[1] = scipy.io.loadmat(filename_label)['Vlabel2']
dataset_A_hold[2] = scipy.io.loadmat(filename_train)['Vdata3']
targets_A_hold[2] = scipy.io.loadmat(filename_label)['Vlabel3']
dataset_A_hold[3] = scipy.io.loadmat(filename_train)['Vdata4']
targets_A_hold[3] = scipy.io.loadmat(filename_label)['Vlabel4']
dataset_A_hold[4] = scipy.io.loadmat(filename_train)['Vdata5']
targets_A_hold[4] = scipy.io.loadmat(filename_label)['Vlabel5']
# %%
dataset_test = scipy.io.loadmat(filename_test)['data']
targets_test = scipy.io.loadmat(filename_test_label)['labels']
X_test = np.reshape(dataset_test, [dataset_test.shape[0], 1, dataset_test.shape[1], dataset_test.shape[2]]).astype(
'float32')
y_test = np.reshape(targets_test, [targets_test.shape[0], 1]).astype('float32')
for epoch1 in range(5): # loop over the dataset multiple times
X_train = np.reshape(dataset_A_train[epoch1],
[dataset_A_train[epoch1].shape[0], 1, dataset_A_train[epoch1].shape[1],
dataset_A_train[epoch1].shape[2]]).astype(
'float32')
y_train = np.reshape(targets_A_train[epoch1], [targets_A_train[epoch1].shape[0], 1]).astype('float32')
X_val = np.reshape(dataset_A_hold[epoch1],
[dataset_A_hold[epoch1].shape[0], 1, dataset_A_hold[epoch1].shape[1],
dataset_A_hold[epoch1].shape[2]]).astype(
'float32')
y_val = np.reshape(targets_A_hold[epoch1], [targets_A_hold[epoch1].shape[0], 1]).astype('float32')
for epoch2 in range(net.epoch):
print("\nEpoch: ", epoch2, epoch1)
print("\nglobal_step: ", global_step)
# 训练网络：输出loss值、global_step用来绘制曲线，训练出的新网络net
loss, global_step, net = train_net.train(net, X_train, y_train, net.batchSize, net.learnRate, global_step, vis)
if vis == 1:
viz.line([loss.item()], [global_step], win='train_loss', update='append', opts={'title':'train_loss'})
# validation and test: 输出预测值，vali_results展示所有指标，test_loss用来绘制test曲线
prediction, valid_results, valid_loss = train_net.valid(net, X_val, y_val)
pred, test_results, test_loss = train_net.valid(net, X_test, y_test)
if vis == 1:
viz.line([test_results], [global_step], win='test_results', update='append', opts={'title':'test_results'})
print('Parameters:["acc", "auc", "recall", "precision","fmeasure"]')
print('validation_results', valid_results)
print('test_results', test_results)