def append_simulations(self, theta: Tensor,
x: Tensor) -> "RestrictionEstimator":
r"""
Store parameters and simulation outputs to use them for training later.
Data ar stored as entries in lists for each type of variable (parameter/data).
Args:
theta: Parameter sets.
x: Simulation outputs.
Returns:
`RestrictionEstimator` object (returned so that this function is chainable).
"""
validate_theta_and_x(theta, x)
if self._valid_or_invalid_criterion == "nan":
label, _, _ = handle_invalid_x(x)
else:
label = self._valid_or_invalid_criterion(x)
label = label.long()
if self._data_round_index:
self._data_round_index.append(self._data_round_index[-1] + 1)
else:
self._data_round_index.append(0)
self._theta_roundwise.append(theta)
self._x_roundwise.append(x)
self._label_roundwise.append(label)
return self
def test_validate_theta_and_x_shapes(
shape_x: Tuple[int], shape_theta: Tuple[int]
) -> None:
x = torch.empty(shape_x)
theta = torch.empty(shape_theta)
validate_theta_and_x(theta, x, training_device="cpu")
def test_validate_theta_and_x_device(device):
# Skip GPU test if not available.
if device == "cuda:0" and not torch.cuda.is_available():
pass
else:
theta = torch.ones((2, 2), dtype=torch.float32).to(device)
x = torch.zeros((2, 10), dtype=torch.float32).to(device)
assert isinstance(
theta, torch.Tensor
), f"{device} based torch.tensor is not an instance of torch.Tensor"
assert theta.dtype == torch.float32, (
f"{device} based torch.tensor(dtype=torch.float32) yields unexpected dtype"
f"{theta.dtype}.")
if device == "cuda:0":
assert not isinstance(
theta, torch.FloatTensor
), f"""{device} based torch.tensor(dtype=torch.float32) must not be
FloatTensor."""
else:
assert isinstance(
theta, torch.FloatTensor
), f"{device} based torch.tensor(dtype=torch.float32) must be FloatTensor."
validate_theta_and_x(theta, x)
with pytest.raises(AssertionError) as _:
validate_theta_and_x(theta, x.to(torch.float64))
plain_ft = torch.FloatTensor((32, 8))
assert (plain_ft.dtype == torch.float32
), "FloatTensor does not expose float32 dtype."
def test_validate_theta_and_x_device(training_device: str, data_device: str) -> None:
theta = torch.empty((1, 1)).to(data_device)
x = torch.empty((1, 1)).to(data_device)
if training_device != data_device:
with pytest.warns(UserWarning):
theta, x = validate_theta_and_x(theta, x, training_device=training_device)
else:
theta, x = validate_theta_and_x(theta, x, training_device=training_device)
assert str(theta.device) == training_device, (
f"Data should have its device converted from '{data_device}' "
f"to training_device '{training_device}'."
)
def test_validate_theta_and_x_gpu():
gpu_if_present = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
theta = torch.ones((32, 8), dtype=torch.float32).to(gpu_if_present)
x = torch.zeros((32, 100), dtype=torch.float32).to(gpu_if_present)
assert isinstance(
theta, torch.Tensor
), "gpu based torch.tensor is not an instance of torch.Tensor"
assert (
theta.dtype == torch.float32
), f"gpu based torch.tensor(dtype=torch.float32) yields unexpected dtype {theta.dtype}."
assert not isinstance(
theta, torch.FloatTensor
), "gpu based torch.tensor(dtype=torch.float32) is FloatTensor, even though it shouldn't be."
validate_theta_and_x(theta, x)
with pytest.raises(AssertionError) as exc:
validate_theta_and_x(theta, x.to(torch.float64))
def test_validate_theta_and_x_cpu():
cpu_device = torch.device("cpu")
theta = torch.ones((32, 8), dtype=torch.float32).to(cpu_device)
x = torch.zeros((32, 100), dtype=torch.float32).to(cpu_device)
plain_ft = torch.FloatTensor((32, 8)) # using an explicit type
assert isinstance(
theta, torch.Tensor
), "cpu based torch.tensor is not an instance of torch.Tensor"
assert (
theta.dtype == torch.float32
), f"cpu based torch.tensor(dtype=torch.float32) yields unexpected dtype {theta.dtype}."
assert isinstance(
theta, torch.FloatTensor
), "cpu based torch.tensor(dtype=torch.float32) is no FloatTensor."
assert plain_ft.dtype == torch.float32, "FloatTensor does not expose float32 dtype."
# test on cpu
validate_theta_and_x(theta, x)
with pytest.raises(AssertionError) as exc:
validate_theta_and_x(theta, x.to(torch.float64))
def test_validate_theta_and_x_type() -> None:
x = torch.empty((1, 1))
theta = torch.empty((1, 1), dtype=int)
with pytest.raises(Exception):
validate_theta_and_x(theta, x, training_device="cpu")
def test_validate_theta_and_x_tensor() -> None:
x = torch.empty((1, 1))
theta = torch.ones((1, 1)).tolist()
with pytest.raises(Exception):
validate_theta_and_x(theta, x, training_device="cpu")