def test_build_ensemble_model(self):
embedding_size = 300
attribute_vector_size = 10
hidden = 0
n_model = 10
dropout_rate_inner = 0.5
dropout_rate_outer = 0.5
dropout_rate_hidden = 0.5
dropout_rate_output = 0.5
padding_final_size = 50
test_model = build_model(embedding_size=embedding_size,
attribute_vector_size=attribute_vector_size,
hidden=hidden,
n_model=n_model,
dropout_rate_inner=dropout_rate_inner,
dropout_rate_outer=dropout_rate_outer,
dropout_rate_hidden=dropout_rate_hidden,
dropout_rate_output=dropout_rate_output,
padding_final_size=padding_final_size)
self.assertEqual(len(test_model.seeds), n_model)
self.assertTrue(isinstance(test_model, EnsembleModel))
self.assertTrue(isinstance(test_model.layers[1], MoleculeConv))
self.assertTrue(isinstance(test_model.layers[2], RandomMask))
self.assertTrue(isinstance(test_model.layers[3], Dense))
def predictor_model(prediction_task="Hf298(kcal/mol)",
embedding_size=512,
attribute_vector_size=None,
depth=2,
add_extra_atom_attribute=True,
add_extra_bond_attribute=True,
differentiate_atom_type=True,
differentiate_bond_type=True,
scale_output=0.05,
padding=False,
padding_final_size=20,
mol_conv_inner_activation='tanh',
mol_conv_outer_activation='softmax',
hidden=0,
hidden_depth=1,
hidden_activation='tanh',
output_activation='linear',
output_size=1,
lr=0.01,
optimizer='adam',
loss='mse',
dropout_rate_inner=0.0,
dropout_rate_outer=0.0,
dropout_rate_hidden=0.0,
dropout_rate_output=0.0,
n_model=None,
freeze_mol_conv=False):
if dropout_rate_inner==0.0 and dropout_rate_outer==0.0 \
and dropout_rate_hidden==0.0 and dropout_rate_output==0.0:
n_model = None
if attribute_vector_size is None:
attribute_vector_size = get_attribute_vector_size(
add_extra_atom_attribute, add_extra_bond_attribute,
differentiate_atom_type, differentiate_bond_type)
model = build_model(embedding_size, attribute_vector_size, depth,
scale_output, padding, mol_conv_inner_activation,
mol_conv_outer_activation, hidden, hidden_depth,
hidden_activation, output_activation, output_size, lr,
optimizer, loss, dropout_rate_inner,
dropout_rate_outer, dropout_rate_hidden,
dropout_rate_output, n_model, padding_final_size,
freeze_mol_conv)
predictor.prediction_task = prediction_task
predictor.model = model
predictor.add_extra_atom_attribute = add_extra_atom_attribute
predictor.add_extra_bond_attribute = add_extra_bond_attribute
predictor.differentiate_bond_type = differentiate_bond_type
predictor.differentiate_atom_type = differentiate_atom_type
predictor.padding = padding
predictor.padding_final_size = padding_final_size
def test_build_model(self):
embedding_size = 300
attribute_vector_size = 10
hidden = 0
test_model = build_model(embedding_size=embedding_size,
attribute_vector_size=attribute_vector_size,
hidden=hidden)
self.assertEqual(len(test_model.layers), 3)
self.assertTrue(isinstance(test_model.layers[1], MoleculeConv))
self.assertTrue(isinstance(test_model.layers[2], Dense))
self.assertEqual(test_model.layers[1].inner_dim,
attribute_vector_size - 1)
self.assertEqual(test_model.layers[1].units, embedding_size)
def test_save_model(self):
embedding_size = 300
attribute_vector_size = 10
hidden = 0
test_model = build_model(embedding_size=embedding_size,
attribute_vector_size=attribute_vector_size,
hidden=hidden)
save_model_folder = os.path.join(os.path.dirname(dde.__file__),
'test_data', 'save_model_test')
if not os.path.exists(save_model_folder):
os.mkdir(save_model_folder)
fpath = os.path.join(save_model_folder, 'model')
save_model(test_model, [1.0], [1.0], 1.0, 1.0, fpath)
import shutil
shutil.rmtree(save_model_folder)
def build_model(self):
"""
This method is intended to provide a way to build default model
"""
self.model = build_model()