def test_independent_reparametrization_sampler_sample_is_repeatable() -> None:
sampler = IndependentReparametrizationSampler(100, _dim_two_gp())
xs = tf.random.uniform([100, 2],
minval=-10.0,
maxval=10.0,
dtype=tf.float64)
npt.assert_allclose(sampler.sample(xs), sampler.sample(xs))
def test_independent_reparametrization_sampler_sample_raises_for_invalid_at_shape(
) -> None:
sampler = IndependentReparametrizationSampler(1,
QuadraticMeanAndRBFKernel())
with pytest.raises(TF_DEBUGGING_ERROR_TYPES):
sampler.sample(tf.constant(0))
def test_independent_reparametrization_sampler_sample_is_continuous() -> None:
sampler = IndependentReparametrizationSampler(100, _dim_two_gp())
xs = tf.random.uniform([100, 2],
minval=-10.0,
maxval=10.0,
dtype=tf.float64)
npt.assert_array_less(
tf.abs(sampler.sample(xs + 1e-20) - sampler.sample(xs)), 1e-20)
def test_independent_reparametrization_sampler_samples_approximate_expected_distribution() -> None:
class ArbitraryButNonTrivialModel(GaussianMarginal):
def predict(self, query_points: TensorType) -> Tuple[TensorType, TensorType]:
latent1 = tf.sin(query_points ** 2) + tf.cos(query_points)
latent2 = tf.sin(query_points) - tf.cos(query_points ** 2)
mean_ = tf.concat([latent1, latent2], axis=-1)
return mean_, tf.sin(mean_)
sample_size = 100
x = tf.linspace([-10.0], [10.0], 20)
model = ArbitraryButNonTrivialModel()
samples = IndependentReparametrizationSampler(sample_size, model).sample(x)
assert samples.shape == [len(x), sample_size, 2]
samples_sorted = tf.sort(samples, axis=-2)
edf = tf.range(1.0, sample_size + 1)[:, None, None] / sample_size
mean, var = model.predict(x)
expected_dist = tfp.distributions.Normal(mean, tf.sqrt(var))
expected_cdf = expected_dist.cdf(tf.transpose(samples_sorted, [1, 0, 2]))
_95_percent_bound = 1.36 / math.sqrt(sample_size)
assert tf.reduce_max(tf.abs(edf - expected_cdf)) < _95_percent_bound
def _build_with_sampler(
self,
dataset: Dataset,
model: ProbabilisticModel,
sampler: IndependentReparametrizationSampler,
) -> AcquisitionFunction:
x = tf.random.uniform([100, 2], minval=-10.0, maxval=10.0, dtype=tf.float64)
samples = sampler.sample(x)
mean, var = model.predict(x)
_assert_kolmogorov_smirnov_95(
tf.linalg.matrix_transpose(samples),
tfp.distributions.Normal(mean[..., None], tf.sqrt(var)[..., None]),
)
return raise_
def test_independent_reparametrization_sampler_samples_approximate_expected_distribution() -> None:
sample_size = 1000
x = tf.random.uniform([100, 2], minval=-10.0, maxval=10.0, dtype=tf.float64)
model = _dim_two_gp()
samples = IndependentReparametrizationSampler(sample_size, model).sample(x) # [N, S, L]
assert samples.shape == [len(x), sample_size, 2]
mean, var = model.predict(x) # [N, L], [N, L]
_assert_kolmogorov_smirnov_95(
tf.linalg.matrix_transpose(samples),
tfp.distributions.Normal(mean[..., None], tf.sqrt(var[..., None])),
)
def test_independent_reparametrization_sampler_samples_are_distinct_for_new_instances(
) -> None:
sampler1 = IndependentReparametrizationSampler(100, _dim_two_gp())
sampler2 = IndependentReparametrizationSampler(100, _dim_two_gp())
xs = tf.random.uniform([100, 2],
minval=-10.0,
maxval=10.0,
dtype=tf.float64)
npt.assert_array_less(1e-9,
tf.abs(sampler2.sample(xs) - sampler1.sample(xs)))
def test_independent_reparametrization_sampler_raises_for_negative_sample_size(
sample_size: int,
) -> None:
with pytest.raises(TF_DEBUGGING_ERROR_TYPES):
IndependentReparametrizationSampler(sample_size, QuadraticMeanAndRBFKernel())
def test_independent_reparametrization_sampler_samples_are_distinct_for_new_instances() -> None:
sampler1 = IndependentReparametrizationSampler(100, QuadraticWithUnitVariance())
sampler2 = IndependentReparametrizationSampler(100, QuadraticWithUnitVariance())
xs = tf.linspace([-10.0], [10.0], 100)
npt.assert_array_less(1e-9, tf.abs(sampler2.sample(xs) - sampler1.sample(xs)))
def test_independent_reparametrization_sampler_sample_is_repeatable() -> None:
sampler = IndependentReparametrizationSampler(100, QuadraticWithUnitVariance())
xs = tf.linspace([-10.0], [10.0], 100)
npt.assert_allclose(sampler.sample(xs), sampler.sample(xs))
def test_independent_reparametrization_sampler_sample_is_continuous() -> None:
sampler = IndependentReparametrizationSampler(100, QuadraticWithUnitVariance())
xs = tf.linspace([-10.0], [10.0], 100)
diff = tf.abs(sampler.sample(xs + 1e-9) - sampler.sample(xs))
npt.assert_array_less(diff, 1e-9)
def test_independent_reparametrization_sampler_sample_raises_for_invalid_at_shape() -> None:
sampler = IndependentReparametrizationSampler(1, QuadraticWithUnitVariance())
with pytest.raises((ValueError, tf.errors.InvalidArgumentError)):
sampler.sample(tf.constant(0))
def test_independent_reparametrization_sampler_raises_for_negative_sample_size(
sample_size: int,
) -> None:
with pytest.raises((ValueError, tf.errors.InvalidArgumentError)):
IndependentReparametrizationSampler(sample_size, QuadraticWithUnitVariance())