def _create_generator(self, *args, **kwargs) -> \
Union[GraphBatchDistanceConvert, GraphBatchGenerator]:
if hasattr(self.graph_converter, 'bond_converter'):
kwargs.update(
{'distance_converter': self.graph_converter.bond_converter})
return GraphBatchDistanceConvert(*args, **kwargs)
return GraphBatchGenerator(*args, **kwargs)
def test_graph_batch_distance_converter(self):
feature = [
np.random.normal(size=(3, 4)),
np.random.normal(size=(2, 4))
]
bond = [np.random.normal(size=(2, )), np.random.normal(size=(1, ))]
glob_features = [
np.random.normal(size=(1, 2)),
np.random.normal(size=(1, 2))
]
index1 = [np.array([0, 1]), np.array([0])]
index2 = [np.array([1, 2]), np.array([1])]
targets = np.random.normal(size=(2, 1))
centers = np.linspace(0, 5, 20)
width = 0.5
gen = GraphBatchDistanceConvert(feature,
bond,
glob_features,
index1,
index2,
targets,
batch_size=2,
distance_converter=GaussianDistance(
centers, width))
data = gen[0]
self.assertListEqual(list(data[0][0].shape), [1, 5, 4])
self.assertListEqual(list(data[0][1].shape), [1, 3, 20])
self.assertListEqual(list(data[0][2].shape), [1, 2, 2])
self.assertListEqual(list(data[0][3].shape), [1, 3])
self.assertListEqual(list(data[0][4].shape), [1, 3])
self.assertListEqual(list(data[1].shape), [1, 2, 1])
def _create_generator(self, *args, **kwargs):
if hasattr(self.graph_convertor, 'bond_convertor'):
kwargs.update(
{'distance_convertor': self.graph_convertor.bond_convertor})
return GraphBatchDistanceConvert(*args, **kwargs)
else:
return GraphBatchGenerator(*args, **kwargs)
def evaluate(test_graphs, test_targets):
"""
Evaluate the test errors using test_graphs and test_targets
Args:
test_graphs (list): list of graphs
test_targets (list): list of target properties
Returns:
mean absolute errors
"""
test_data = model.graph_converter.get_flat_data(test_graphs, test_targets)
gen = GraphBatchDistanceConvert(
*test_data,
distance_converter=model.graph_converter.bond_converter,
batch_size=128)
preds = []
trues = []
for i in range(len(gen)):
d = gen[i]
preds.extend(model.predict(d[0]).ravel().tolist())
trues.extend(d[1].ravel().tolist())
return np.mean(np.abs(np.array(preds) - np.array(trues)))
def test_graph_to_inputs_and_class_generator(self):
graphs = [index_rep_from_structure(i) for i in self.structures] * 4
mp_ids = ['mp-19017', 'mp-2998'] * 5
targets = [0.1, 0.2] * 5
out = graph_to_inputs(mp_ids, graphs, targets)
self.assertEqual(len(out), 7)
gen = GraphBatchDistanceConvert(*out[:-1],
batch_size=2,
distance_converter=GaussianDistance())
data = gen[0]
x = data[0]
y = data[1]
# only one graph, therefore the batch dimension is 1
self.assertListEqual([i.shape[0] for i in x], [1] * len(x))
# atom is 1*N where N is the total number of atoms
self.assertEqual(len(x[0].shape), 2)
# bond is 1*M*G
self.assertEqual(len(x[1].shape), 3)
# global is 1*U*K
self.assertEqual(len(x[2].shape), 3)
self.assertListEqual([len(x[i].shape) for i in range(3, 7)], [2] * 4)
# target is 1*2(crystal)*1(target dimension)
self.assertListEqual(list(y.shape), [1, 2, 1])
def _create_generator(self, *args, **kwargs):
if self.distance_convertor is not None:
kwargs.update({'distance_convertor': self.distance_convertor})
return GraphBatchDistanceConvert(*args, **kwargs)
else:
return GraphBatchGenerator(*args, **kwargs)