def test_subset_with_no_seed():
dataset = list(np.random.rand(100))
subset1 = get_random_subset(dataset, 10)
subset2 = get_random_subset(dataset, 10)
assert subset1 != subset2
def test_subset_with_random_seed():
dataset = list(np.random.rand(100))
subset1 = get_random_subset(dataset, 10, seed=33)
subset2 = get_random_subset(dataset, 10, seed=33)
subset3 = get_random_subset(dataset, 10, seed=43)
assert subset1 == subset2
assert subset1 != subset3
def test_subset():
dataset = list(np.random.rand(100))
subset = get_random_subset(dataset, 10)
for s in subset:
assert s in dataset
def __init__(self, training_set: List[str],
chemnet_model_filename='ChemNet_v0.13_pretrained.h5',
sample_size=10000) -> None:
"""
Args:
training_set: molecules from the training set
chemnet_model_filename: name of the file for trained ChemNet model.
Must be present in the 'fcd' package, since it will be loaded directly from there.
sample_size: how many molecules to generate the distribution statistics from (both reference data and model)
"""
self.chemnet_model_filename = chemnet_model_filename
self.sample_size = sample_size
super().__init__(name='Frechet ChemNet Distance', number_samples=self.sample_size)
self.reference_molecules = get_random_subset(training_set, self.sample_size, seed=42)
def __init__(self, number_samples: int, training_set: List[str]) -> None:
"""
Args:
number_samples: number of samples to generate from the model
training_set: molecules from the training set
"""
super().__init__(name='KL divergence', number_samples=number_samples)
self.training_set_molecules = canonicalize_list(
get_random_subset(training_set, self.number_samples, seed=42),
include_stereocenters=False)
self.pc_descriptor_subset = [
'BertzCT', 'MolLogP', 'MolWt', 'TPSA', 'NumHAcceptors',
'NumHDonors', 'NumRotatableBonds', 'NumAliphaticRings',
'NumAromaticRings'
]
def test_subset_if_dataset_too_small():
dataset = list(np.random.rand(100))
with pytest.raises(Exception):
get_random_subset(dataset, 1000)
