def test_extend_adj(adj):
adj_extended = extend_adj(adj, out_size=6)
assert adj_extended.shape == (batchsize, 6, 6)
assert adj_extended.dtype == adj.dtype
assert numpy.array_equal(adj_extended[:, :num_node, :num_node], adj)
assert numpy.alltrue(adj_extended[:, num_node:, :] == 0)
assert numpy.alltrue(adj_extended[:, :, num_node:] == 0)
def test_extend_adj(adj):
adj_extended = extend_adj(adj, out_size=6)
assert adj_extended.shape == (batchsize, 6, 6)
assert adj_extended.dtype == adj.dtype
assert numpy.array_equal(adj_extended[:, :num_node, :num_node], adj)
assert numpy.alltrue(adj_extended[:, num_node:, :] == 0)
assert numpy.alltrue(adj_extended[:, :, num_node:] == 0)
def test_forward_cpu_input_size_invariant(model, data):
atom_data, adj_data = data[0], data[1]
is_real_node = numpy.ones(atom_data.shape, dtype=numpy.float32)
y_actual = cuda.to_cpu(model(atom_data, adj_data, is_real_node).data)
atom_data_ex = extend_node(atom_data, out_size=8)
adj_data_ex = extend_adj(adj_data, out_size=8)
is_real_node_ex = extend_node(is_real_node, out_size=8)
y_actual_ex = cuda.to_cpu(
model(atom_data_ex, adj_data_ex, is_real_node_ex).data)
assert numpy.allclose(y_actual, y_actual_ex, rtol=1e-5, atol=1e-6)
def test_forward_cpu_input_size_invariant(model, data):
atom_data, adj_data = data[0], data[1]
is_real_node = numpy.ones(atom_data.shape, dtype=numpy.float32)
y_actual = cuda.to_cpu(model(atom_data, adj_data, is_real_node).data)
atom_data_ex = extend_node(atom_data, out_size=8)
adj_data_ex = extend_adj(adj_data, out_size=8)
is_real_node_ex = extend_node(is_real_node, out_size=8)
y_actual_ex = cuda.to_cpu(model(
atom_data_ex, adj_data_ex, is_real_node_ex).data)
assert numpy.allclose(y_actual, y_actual_ex, rtol=1e-5, atol=1e-6)
def test_forward_cpu_input_size_invariant(model, data):
# This RSGCN uses dropout, so we need to forward with test mode
# to remove stochastic calculation.
atom_data, adj_data = data[0], data[1]
with chainer.using_config('train', False):
y_actual = cuda.to_cpu(model(atom_data, adj_data).data)
# Set bigger size than original `atom_size`.
atom_data_ex = extend_node(atom_data, out_size=8)
adj_data_ex = extend_adj(adj_data, out_size=8)
# print('size', atom_data.shape, adj_data.shape,
# atom_data_ex.shape, adj_data_ex.shape)
with chainer.using_config('train', False):
y_actual_ex = cuda.to_cpu(model(atom_data_ex, adj_data_ex).data)
assert numpy.allclose(y_actual, y_actual_ex, rtol=1.e-4, atol=1.e-5)
def test_forward_cpu_input_size_invariant(model, data):
# This RSGCN uses dropout, so we need to forward with test mode
# to remove stochastic calculation.
atom_data, adj_data = data[0], data[1]
with chainer.using_config('train', False):
y_actual = cuda.to_cpu(model(atom_data, adj_data).data)
# Set bigger size than original `atom_size`.
atom_data_ex = extend_node(atom_data, out_size=8)
adj_data_ex = extend_adj(adj_data, out_size=8)
# print('size', atom_data.shape, adj_data.shape,
# atom_data_ex.shape, adj_data_ex.shape)
with chainer.using_config('train', False):
y_actual_ex = cuda.to_cpu(model(
atom_data_ex, adj_data_ex).data)
assert numpy.allclose(y_actual, y_actual_ex, rtol=1.e-4, atol=1.e-5)
def test_padding(self, sample_molecule_2):
adj = common.construct_discrete_edge_matrix(sample_molecule_2, 8)
assert adj.shape == (4, 8, 8)
expect = extend_adj(self.expect_adj, out_size=8, axis=[-1, -2])
numpy.testing.assert_equal(adj, expect)
def test_padding(self, sample_molecule_2):
adj = common.construct_discrete_edge_matrix(sample_molecule_2, 8)
assert adj.shape == (4, 8, 8)
expect = extend_adj(self.expect_adj, out_size=8, axis=[-1, -2])
numpy.testing.assert_equal(adj, expect)
def test_extend_adj_assert_raises():
with pytest.raises(ValueError):
extend_adj(adj_3d, out_size=1)
def test_extend_adj_assert_raises():
with pytest.raises(ValueError):
extend_adj(adj_3d, out_size=1)
