type=str,
default='result.txt')
parser.add_argument('--smiles_data', help='smiles data', type=str)
parser.add_argument('--vocab_from',
help='the file where vocab is extracted from',
type=str)
parser.add_argument('--lr', help='learning rate', type=float, default=0.0001)
args = parser.parse_args()
#extact vocab and char
char, vocab = extract_vocab(args.vocab_from, args.seq_length)
vocab_size = len(char)
#model and restore model parapmeters
model = CVAE(vocab_size, args)
model.restore(args.save_file)
print('Number of parameters : ',
np.sum([np.prod(v.shape) for v in tf.trainable_variables()]))
#target property to numpy array
start_codon = np.array(
[np.array(list(map(vocab.get, 'X'))) for _ in range(args.batch_size)])
#generate smiles
smiles = []
for _ in range(args.num_iteration):
generated = model.sample(start_codon, args.seq_length)
smiles += [
convert_to_smiles(generated[i], char) for i in range(len(generated))
]
if not os.path.isdir(args.save_dir):
os.mkdir(args.save_dir)
#divide data into training and test set
num_train_data = int(len(molecules_input) * 0.75)
train_molecules_input = molecules_input[0:num_train_data]
test_molecules_input = molecules_input[num_train_data:-1]
train_molecules_output = molecules_output[0:num_train_data]
test_molecules_output = molecules_output[num_train_data:-1]
train_length = length[0:num_train_data]
test_length = length[num_train_data:-1]
model = CVAE(vocab_size, args)
if args.pretrained is not None: model.restore(args.pretrained)
print('Number of parameters : ',
np.sum([np.prod(v.shape) for v in tf.trainable_variables()]))
for epoch in range(args.num_epochs):
st = time.time()
# 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)