def test_validity_correct_molecules():
smiles_1 = 'O'
smiles_2 = 'C'
smiles_3 = 'CC(ONONOC)CCCc1ccccc1'
assert is_valid(smiles_1)
assert is_valid(smiles_2)
assert is_valid(smiles_3)
def sample_valid_molecules(model: DistributionMatchingGenerator, number_molecules: int, max_tries=10) -> List[str]:
"""
Sample from the given generator until the desired number of valid molecules
has been sampled (i.e., ignore invalid molecules).
Args:
model: model to sample from
number_molecules: number of valid molecules to generate
max_tries: determines the maximum number N of samples to draw, N = number_molecules * max_tries
Returns:
A list of number_molecules valid molecules. If this was not possible with the given max_tries, the list may be shorter.
"""
max_samples = max_tries * number_molecules
number_already_sampled = 0
valid_molecules: List[str] = []
while len(valid_molecules) < number_molecules and number_already_sampled < max_samples:
remaining_to_sample = number_molecules - len(valid_molecules)
samples = model.generate(remaining_to_sample)
number_already_sampled += remaining_to_sample
valid_molecules += [m for m in samples if is_valid(m)]
return valid_molecules
def assess_model(
self, model: DistributionMatchingGenerator
) -> DistributionLearningBenchmarkResult:
start_time = time.time()
molecules = model.generate(number_samples=self.number_samples)
end_time = time.time()
if len(molecules) != self.number_samples:
raise Exception(
"The model did not generate the correct number of molecules")
number_valid = sum(1 if is_valid(smiles) else 0
for smiles in molecules)
validity_ratio = number_valid / self.number_samples
metadata = {
"number_samples": self.number_samples,
"number_valid": number_valid
}
return DistributionLearningBenchmarkResult(
benchmark_name=self.name,
score=validity_ratio,
sampling_time=end_time - start_time,
metadata=metadata,
)
def test_validity_incorrect_syntax():
smiles = 'CCCincorrectsyntaxCCC'
assert not is_valid(smiles)
def test_validity_empty_molecule():
smiles = ''
assert not is_valid(smiles)
def test_validity_incorrect_valence():
smiles = 'CCC(CC)(CC)(=O)CCC'
assert not is_valid(smiles)
