def encode(self, chunk, use_random=False):
"""
Args:
chunk: a list of `n` strings, each being a SMILES.
Returns:
A numpy array of dtype np.float32, of shape (n, latent_dim)
Note: Each row should be the *mean* of the latent space distrubtion rather than a sampled point from that distribution.
(It can be anythin as long as it fits what self.decode expects)
"""
cfg_tree_list = []
for smiles in chunk:
ts = parser.parse(smiles, self.grammar)
assert isinstance(ts, list) and len(ts) == 1
n = AnnotatedTree2MolTree(ts[0])
cfg_tree_list.append(n)
if type(chunk[0]) is str:
cfg_tree_list = parse(chunk, self.grammar)
else:
cfg_tree_list = chunk
onehot, _ = batch_make_att_masks(cfg_tree_list,
self.tree_decoder,
self.onehot_walker,
dtype=np.float32)
x_inputs = np.transpose(onehot, [0, 2, 1])
x_inputs = paddle.to_tensor(x_inputs)
z_mean, _ = self.ae.encoder(x_inputs)
return z_mean.numpy()
def parse_single(smiles, grammar):
"""
tbd
"""
ts = parser.parse(smiles, grammar)
assert isinstance(ts, list) and len(ts) == 1
n = AnnotatedTree2MolTree(ts[0])
return n
def parse_smiles_with_cfg(smiles_file, grammar_file):
grammar = parser.Grammar(grammar_file)
cfg_tree_list = []
with open(smiles_file, 'r') as f:
for row in tqdm(f):
smiles = row.strip()
ts = parser.parse(smiles, grammar)
assert isinstance(ts, list) and len(ts) == 1
n = AnnotatedTree2MolTree(ts[0])
cfg_tree_list.append(n)
return cfg_tree_list
def process_chunk(smiles_list):
grammar = parser.Grammar(cmd_args.grammar_file)
cfg_tree_list = []
for smiles in smiles_list:
ts = parser.parse(smiles, grammar)
assert isinstance(ts, list) and len(ts) == 1
n = AnnotatedTree2MolTree(ts[0])
cfg_tree_list.append(n)
walker = OnehotBuilder()
tree_decoder = create_tree_decoder()
onehot, masks = batch_make_att_masks(cfg_tree_list, tree_decoder, walker, dtype=np.byte)
return (onehot, masks)
def parse_smiles_with_cfg(smiles_file, grammar_file):
grammar = parser.Grammar(cmd_args.grammar_file)
smiles_list = []
cfg_tree_list = []
annotated_trees = []
with open(smiles_file, 'r') as f:
for row in tqdm(f):
smiles = row.strip()
smiles_list.append(smiles)
ts = parser.parse(smiles, grammar)
assert isinstance(ts, list) and len(ts) == 1
annotated_trees.append(ts[0])
n = AnnotatedTree2MolTree(ts[0])
cfg_tree_list.append(n)
st = get_smiles_from_tree(n)
assert st == smiles
return (smiles_list, cfg_tree_list, annotated_trees)
h = self.w1(flatten)
h = F.relu(h)
z_mean = self.mean_w(h)
z_log_var = self.log_var_w(h)
return (z_mean, z_log_var)
if __name__ == '__main__':
smiles_list = ['N\SCPP#IOS', 'CP\P', 'PINI']
cfg_trees = []
cfg_onehots = []
grammar = parser.Grammar(cmd_args.grammar_file)
for smiles in smiles_list:
ts = parser.parse(smiles, grammar)
assert isinstance(ts, list) and len(ts) == 1
n = AnnotatedTree2MolTree(ts[0])
cfg_trees.append(n)
cfg_onehots.append(AnnotatedTree2Onehot(ts[0], 50))
cfg_onehots = np.stack(cfg_onehots, axis=0)
encoder = CNNEncoder(max_len=50, latent_dim=64)
if cmd_args.mode == 'gpu':
encoder.cuda()
z = encoder(cfg_onehots)
print(z[0].size())