def trans_mode(mode='normal'):
global criterion
global train_set
global test_set
if mode == 'one-hot':
label_file = basic_dir + 'one-hot-label.csv'
train_set = MusicDataThree(data_file,
label_file,
start=train_start,
total=train_end)
test_set = MusicDataThree(data_file,
label_file,
start=test_start,
total=test_end)
criterion = nn.CrossEntropyLoss()
set_record_file('record-modeltest.txt')
print('Test of deep convolutional network')
data_dir = basic_dir + 'raw-data-v5/'
data_file = basic_dir + 'music-data-v5.csv'
label_file = basic_dir + 'labels-v5.csv'
# net_name = 'shallownet'
net_name = '0-justreducelr'
# net_name = 'simplecnn-freq'
model_name = net_name + '.pt'
net = Justreducelr_0() # deep convolutional network
net.load_state_dict(torch.load(model_name))
net.eval()
# test_set = MusicDataThree(data_file=data_file, label_file=label_file, start=2560, total=3219)
test_set = MusicDataThree(data_file=data_file,
label_file=label_file,
start=2560,
total=3219)
test(net, net_name=net_name, dataset=test_set)
print('Test of shallow convolutional network')
data_file = basic_dir + 'music-data-v5.csv'
label_file = basic_dir + 'labels-v5.csv'
net = ShallowNet()
net_name = 'shallownet'
model_name = net_name + '.pt'
net.load_state_dict(torch.load(model_name))
net.eval()
test_set = MusicDataThree(data_file=data_file,
label_file=label_file,
start=2560,
total=3219)
from train_test import train, test
from train_test import Ininorm_5_6 as Net
from MusicDataset import MusicDataThree, IniNorm
# time.sleep(3600*4)
print('start')
model_path = '6-5+ininorm.pt'
print('start')
model_path = '6-5+ininorm.pt'
tsfm = IniNorm()
train_set = MusicDataThree(transform=tsfm)
test_set = MusicDataThree(transform=tsfm)
net = Net()
net = train(net, model_path=model_path, dataset=train_set)
test(net, dataset=test_set)
from train_test import train, test, set_record_file
from train_test import Coon_0_2 as Net
from MusicDataset import MusicDataThree, ExpNorm
# time.sleep(3600*4)
print('start')
set_record_file('record-expoon.txt')
net_name = '2-0+expcoon'
model_path = net_name + '.pt'
tsfm = ExpNorm()
train_set = MusicDataThree(transform=tsfm, start=0, total=1)
test_set = MusicDataThree(transform=tsfm, start=0, total=1)
net = Net()
net = train(net, model_path=model_path, dataset=train_set)
test(net, net_name, dataset=test_set)
return self.classifier(x).view(-1, 18)
def weights_init(m):
if isinstance(m, nn.Conv2d):
nn.init.normal_(m.weight.data, mean=0, std=10)
nn.init.normal_(m.bias.data, mean=0, std=10)
net = Justreducelr_0()
net.apply(weights_init)
# criterion = nn.CrossEntropyLoss()
# criterion = nn.BCELoss() # ##################################################### here
train_set = MusicDataThree(data_file,
label_file,
start=train_start,
total=train_end)
test_set = MusicDataThree(data_file,
label_file,
start=test_start,
total=test_end)
criterion = nn.BCEWithLogitsLoss()
def set_record_file(file_name):
global record_file
record_file = file_name
def trans_mode(mode='normal'):
global criterion
second = F.relu(self.pool(self.conv3(second)))
second = F.relu(self.pool(self.conv4(second)))
# for i in range(3):
# second = F.relu(self.conv5(second))
# second = self.conv6(second)
second = second.view(batch_size, -1)
y = self.sigmoid(first + second)
return y
net = Coon_0_2()
# criterion = nn.BCELoss() # ##################################################### here
train_set = MusicDataThree(data_file,
label_file,
start=train_start,
total=train_end,
mode='one-hot')
test_set = MusicDataThree(data_file,
label_file,
start=test_start,
total=test_end,
mode='one-hot')
# criterion = nn.MSELoss()
criterion = nn.CrossEntropyLoss()
def train(net=net,
criterion=criterion,
model_path='tmp.pt',
dataset=train_set,