def modelTrain(x_data,y_data):
# now = datetime.utcnow().strftime("%Y%m%d")
# model_dir = "model/{}/".format(now)
model_dir = "Model/Regression/"+save_date+"/"
name = model_dir + gas_type + "_reg"
try:
if not (os.path.isdir(model_dir)):
os.makedirs(os.path.join(model_dir))
except OSError as e:
if e.errno != errno.EEXIST:
raise
start = timeit.default_timer()
before = 0
x_data = np.reshape(x_data, (-1, 1, 6))
y_data = np.reshape(y_data, (-1, 1))
x_data = torch.Tensor(x_data).float()
y_data = torch.Tensor(y_data).float()
input = len(x_data[0])
hidden = 100
hidden2 = 500
hidden3 = 100
hidden4 = 300
hidden5 = 500
output = 1
epoch = 20000
learning_rate = 0.001
batch_size = 10
cnn = CNN(input, output, [1,30, 30, -1, 10], [hidden, hidden2, hidden3, hidden4, hidden5], kernel_size=2, stride=2, padding=1, batch_size = batch_size, softmax=False)
for step, loss in cnn.learn(x_data, y_data, epoch=epoch, learning_rate=learning_rate, loss_func=MSELoss):
if step % 1000 == 0:
gap = loss.tolist() - before
before = loss.tolist()
pick = timeit.default_timer()
time_gap = pick-start
print('step : %d, loss : %.10f, loss gap = %.10f, seconds = %d' %(step,loss.tolist(),gap,time_gap))
if loss==0:
gap = loss.tolist() - before
pick = timeit.default_timer()
time_gap = pick - start
break
print('step : %d, loss : %.10f, loss gap = %.10f, seconds = %d' % (step, loss.tolist(), gap, time_gap))
save_model(cnn, name+'.pkl')
from datetime import datetime
from datetime import datetime