model.assign_lr(learning_rate * (decay_rate ** epoch))
train_loss = []
test_loss = []
st = time.time()
for iteration in range(int(len(train_molecules)/batch_size)):
n = np.random.randint(len(train_molecules), size = batch_size)
x = np.array([train_molecules[i] for i in n])
l = np.array([train_length[i] for i in n])
c = np.array([train_labels[i] for i in n])
cost = model.train(x, l, c)
train_loss.append(cost)
for iteration in range(int(len(test_molecules)/batch_size)):
n = np.random.randint(len(test_molecules), size = batch_size)
x = np.array([test_molecules[i] for i in n])
l = np.array([test_length[i] for i in n])
c = np.array([test_labels[i] for i in n])
cost = model.test(x, l, c)
test_loss.append(cost)
train_loss = np.mean(np.array(train_loss))
test_loss = np.mean(np.array(test_loss))
end = time.time()
if epoch==0:
print ('epoch\ttrain_loss\ttest_loss\ttime (s)')
print ("%s\t%.3f\t%.3f\t%.3f" %(epoch, train_loss, test_loss, end-st))
ckpt_path = 'save/model.ckpt'
model.save(ckpt_path, epoch)
# Learning rate scheduling
#model.assign_lr(learning_rate * (decay_rate ** epoch))
train_loss = []
test_loss = []
st = time.time()
for iteration in range(len(train_molecules_input) // args.batch_size):
n = np.random.randint(len(train_molecules_input), size=args.batch_size)
x = np.array([train_molecules_input[i] for i in n])
y = np.array([train_molecules_output[i] for i in n])
l = np.array([train_length[i] for i in n])
cost = model.train(x, y, l)
train_loss.append(cost)
for iteration in range(len(test_molecules_input) // args.batch_size):
n = np.random.randint(len(test_molecules_input), size=args.batch_size)
x = np.array([test_molecules_input[i] for i in n])
y = np.array([test_molecules_output[i] for i in n])
l = np.array([test_length[i] for i in n])
cost = model.test(x, y, l)
test_loss.append(cost)
train_loss = np.mean(np.array(train_loss))
test_loss = np.mean(np.array(test_loss))
end = time.time()
if epoch == 0:
print('epoch\ttrain_loss\ttest_loss\ttime (s)')
print("%s\t%.3f\t%.3f\t%.3f" % (epoch, train_loss, test_loss, end - st))
ckpt_path = args.save_dir + '/model_' + str(epoch) + '.ckpt'
model.save(ckpt_path, epoch)