def setUp(self):
self.temp_file = TempSMILESFile()
self.fh = self.temp_file.open()
# See `test_data.py` for data set test cases.
self.dataset = SMILESDataset(self.fh.name)
self.vocab = SMILESVocabulary(self.dataset, need_corpus=True)
class SMILESConsecutiveSamplerTestCase(unittest.TestCase):
def setUp(self):
self.smiles_string = 'CCc1c[n+]2ccc3c4ccccc4[nH]c3c2cc1'
with TempSMILESFile(smiles_strings=self.smiles_string) as temp_fh:
dataset = SMILESDataset(temp_fh.file_handler.name)
self.vocabulary = SMILESVocabulary(dataset, need_corpus=True)
def test_sampling_with_padding(self):
smiles_string = Token.tokenize(Token.augment(self.smiles_string))
n_steps = 20
sampler = SMILESConsecutiveSampler(self.vocabulary.corpus,
n_steps=n_steps)
step_i = 0
for sample in sampler:
input_s = ''.join(smiles_string[step_i:step_i + n_steps])
output_s = ''.join(smiles_string[step_i + 1:step_i + n_steps + 1])
if sample.valid_length < n_steps:
input_s += Token.PAD * (n_steps - sample.valid_length - 1)
output_s += Token.PAD * (n_steps - sample.valid_length)
self.assertEqual(
input_s,
''.join(self.vocabulary.get_tokens(sample.inputs)),
)
self.assertEqual(
output_s,
''.join(self.vocabulary.get_tokens(sample.outputs)),
)
step_i += n_steps
def test_sampling_without_padding(self):
tokens = Token.tokenize(Token.augment(self.smiles_string))
n_steps = len(tokens) - 1
sampler = SMILESConsecutiveSampler(self.vocabulary.corpus,
n_steps=n_steps)
step_i = 0
for n_samples, sample in enumerate(sampler, start=1):
self.assertListEqual(
tokens[step_i:step_i + n_steps],
self.vocabulary.get_tokens(sample.inputs),
)
self.assertListEqual(
tokens[step_i + 1:step_i + n_steps + 1],
self.vocabulary.get_tokens(sample.outputs),
)
self.assertEqual(n_steps, sample.valid_length)
step_i += n_steps
self.assertEqual(n_samples, 1)
class SMILESVocabularyTestCase(unittest.TestCase):
def setUp(self):
self.temp_file = TempSMILESFile()
self.fh = self.temp_file.open()
# See `test_data.py` for data set test cases.
self.dataset = SMILESDataset(self.fh.name)
self.vocab = SMILESVocabulary(self.dataset, need_corpus=True)
def test_tokens_and_idx(self):
self.assertSequenceEqual(
# Tokenize the entire dataset to get a set of unique tokens.
sorted(
set(
Token.tokenize(
self.temp_file.smiles_strings.replace('\n', '')))),
# The temporary file is not augmented by the special tokens.
sorted(set(self.vocab.token_to_idx) - Token.SPECIAL),
)
self.assertSequenceEqual(
sorted(
set(self.vocab.token_to_idx)
# Pad and unknown tokens does not appear in the original set.
- {Token.PAD, Token.UNK}),
sorted(set(self.vocab.token_freqs)),
)
def test_corpus(self):
# Original SMILES list without padded special tokens.
smiles_list = self.temp_file.smiles_strings.split('\n')
self.assertEqual(len(self.vocab.corpus), len(smiles_list))
for idx, tokens in zip(self.vocab.corpus, smiles_list):
# Add special tokens in order to correspond to the loaded corpus
# for data sampling and model fitting.
tokens = Token.augment(tokens)
# Test id-to-token mapping.
self.assertEqual(
''.join(self.vocab.get_tokens(idx)),
tokens,
)
# Test token-to-id mapping.
self.assertListEqual(idx, self.vocab[Token.tokenize(tokens)])
def test_contains(self):
self.assertNotIn(Token.UNK, self.vocab)
all_tokens = Token.get_all_tokens()
for token in self.vocab:
if len(token) == 1 and token.islower():
token = token.upper()
self.assertIn(token, all_tokens)
def tearDown(self):
self.fh.close()
def setUp(self):
temp_file = TempSMILESFile(
tempfile_kwargs={'prefix': 'softmax_sampler'})
self.fh = temp_file.open()
dataset = SMILESDataset(self.fh.name)
self.vocabulary = SMILESVocabulary(dataset, need_corpus=True)
self.model = SMILESRNN(len(self.vocabulary))
self.predictor = SoftmaxSearch(self.model, self.vocabulary)
def setUp(self):
self.temp_file = TempSMILESFile(
tempfile_kwargs={'prefix': 'dataloader'})
self.fh = self.temp_file.open()
dataset = SMILESDataset(self.fh.name)
vocabulary = SMILESVocabulary(dataset=dataset, need_corpus=True)
self.dataloader = SMILESBatchColumnSampler(
corpus=vocabulary.corpus,
batch_size=2,
n_steps=4,
shuffle=True,
)
def setUp(self):
temp_file = TempSMILESFile(tempfile_kwargs={'prefix': 'model'})
self.fh = temp_file.open()
dataset = SMILESDataset(self.fh.name)
self.vocabulary = SMILESVocabulary(dataset, need_corpus=True)
self.batch_sampler = SMILESBatchColumnSampler(
corpus=self.vocabulary.corpus,
batch_size=3,
n_steps=8,
)
self.n_rnn_layers = 1 # Used in output/state shape testing.
self.n_rnn_units = 32 # Used in output/state shape testing.
self.model = SMILESRNN(len(self.vocabulary),
use_one_hot=False,
embedding_dim=4,
embedding_dropout=0.25,
embedding_dropout_axes=0,
embedding_init=mx.init.Uniform(),
embedding_prefix='embedding_',
rnn='lstm',
rnn_n_layers=self.n_rnn_layers,
rnn_n_units=self.n_rnn_units,
rnn_i2h_init='xavier_normal',
rnn_h2h_init='orthogonal_normal',
rnn_reinit_state=True,
rnn_detach_state=False,
rnn_state_init=mx.nd.random.uniform,
rnn_dropout=0.0,
rnn_prefix='encoder_',
dense_n_layers=2,
dense_n_units=32,
dense_activation='relu',
dense_dropout=0.5,
dense_init=mx.init.Xavier(),
dense_prefix='decoder_',
dtype='float32',
prefix='model_')
def setUp(self):
self.smiles_string = 'CCc1c[n+]2ccc3c4ccccc4[nH]c3c2cc1'
with TempSMILESFile(smiles_strings=self.smiles_string) as temp_fh:
dataset = SMILESDataset(temp_fh.file_handler.name)
self.vocabulary = SMILESVocabulary(dataset, need_corpus=True)