def test_sparse_variational_optimize_with_defaults() -> None:
x_observed = np.linspace(0, 100, 100).reshape((-1, 1))
y_observed = _3x_plus_gaussian_noise(x_observed)
data = x_observed, y_observed
dataset = Dataset(*data)
optimizer = create_optimizer(tf.optimizers.Adam(), dict(max_iter=20))
model = SparseVariational(svgp_model(x_observed, y_observed), optimizer=optimizer)
loss = model.model.training_loss(data)
model.optimize(dataset)
assert model.model.training_loss(data) < loss
def test_sparse_variational_optimize(batcher: DatasetTransformer, compile: bool) -> None:
x_observed = np.linspace(0, 100, 100).reshape((-1, 1))
y_observed = _3x_plus_gaussian_noise(x_observed)
data = x_observed, y_observed
dataset = Dataset(*data)
optimizer = create_optimizer(
tf.optimizers.Adam(),
dict(max_iter=10, batch_size=10, dataset_builder=batcher, compile=compile),
)
model = SparseVariational(svgp_model(x_observed, y_observed), optimizer=optimizer)
loss = model.model.training_loss(data)
model.optimize(dataset)
assert model.model.training_loss(data) < loss
def test_sparse_variational_update_updates_num_data() -> None:
model = SparseVariational(
svgp_model(tf.zeros([1, 4]), tf.zeros([1, 1])),
)
model.update(Dataset(tf.zeros([5, 4]), tf.zeros([5, 1])))
assert model.model.num_data == 5
def test_sparse_variational_update_raises_for_invalid_shapes(new_data: Dataset) -> None:
model = SparseVariational(
svgp_model(tf.zeros([1, 4]), tf.zeros([1, 1])),
)
with pytest.raises(ValueError):
model.update(new_data)
def test_sparse_variational_model_attribute() -> None:
model = svgp_model(*mock_data())
sv = SparseVariational(model)
assert sv.model is model