def testSamples(self):
for samples_shape in ([2], [2, 4], [4, 2, 4]):
input_ = random_samples(samples_shape)
input_ph = tf.compat.v1.placeholder_with_default(
input=input_, shape=samples_shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph)
self.assertAllClose(input_ph, self.evaluate(dist.samples))
invalid_sample = 0
with self.assertRaises(ValueError):
dist = empirical.Empirical(samples=invalid_sample)
def testSamples(self):
for samples_shape in ([4, 2, 4], [4, 2, 2, 4]):
input_ = random_samples(samples_shape)
input_ph = tf.placeholder_with_default(
input=input_,
shape=samples_shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph, event_ndims=2)
self.assertAllClose(input_ph, self.evaluate(dist.samples))
invalid_samples = [0, [0, 1], [[0, 1], [1, 2]]]
for samples in invalid_samples:
with self.assertRaises(ValueError):
dist = empirical.Empirical(samples=samples, event_ndims=2)
def testPmfWithBatch(self):
sample = [[[0, 0], [0, 1], [0, 1], [1, 1]],
[[0, 10], [10, 10], [10, 20], [20, 20]]]
events = [
[0, 1],
[[0, 1], [10, 10]],
[[[0, 0], [0, 10]],
[[0, 1], [10, 20]]],
[[0], [10]],
[[[0, 1]], [[10, 20]]]
]
expected_pmfs = [
[0.5, 0],
[0.5, 0.25],
[[0.25, 0.25], [0.5, 0.25]],
[0.25, 0.25],
[[0.5, 0], [0, 0.25]]
]
for event, expected_pmf in zip(events, expected_pmfs):
input_ = tf.convert_to_tensor(value=sample, dtype=np.float32)
input_ph = tf.compat.v1.placeholder_with_default(
input=input_, shape=input_.shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph, event_ndims=1)
self.assertAllClose(self.evaluate(dist.prob(event)),
expected_pmf)
self.assertAllClose(self.evaluate(dist.log_prob(event)),
np.log(expected_pmf))
def testShapes(self):
for samples_shape in ([2], [2, 4], [4, 2, 4]):
input_ = random_samples(samples_shape)
input_ph = tf.compat.v1.placeholder_with_default(
input=input_, shape=samples_shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph)
self.assertAllEqual(samples_shape[:-1],
self.evaluate(dist.batch_shape_tensor()))
self.assertAllEqual([],
self.evaluate(dist.event_shape_tensor()))
def testLogProbAfterSlice(self):
samples = np.random.randn(6, 5, 4)
dist = empirical.Empirical(samples=samples, event_ndims=1)
self.assertAllEqual((6,), dist.batch_shape)
self.assertAllEqual((4,), dist.event_shape)
sliced_dist = dist[:, tf.newaxis]
samples = self.evaluate(dist.sample())
self.assertAllEqual((6, 4), samples.shape)
lp, sliced_lp = self.evaluate([
dist.log_prob(samples), sliced_dist.log_prob(samples[:, tf.newaxis])])
self.assertAllEqual(lp[:, tf.newaxis], sliced_lp)
def testEntropy(self):
sample = [[[0, 0], [0, 1]], [[0, 1], [1, 2]], [[0, 1], [1, 2]],
[[0, 2], [2, 4]]]
expected_entropy = entropy([0.25, 0.5, 0.25])
input_ = tf.convert_to_tensor(sample, dtype=np.float64)
input_ph = tf.placeholder_with_default(
input=input_, shape=input_.shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph, event_ndims=2)
self.assertAllClose(self.evaluate(dist.entropy()), expected_entropy)
def testMode(self):
sample = [[[0, 0], [0, 1]], [[0, 1], [1, 2]], [[0, 1], [1, 2]],
[[0, 2], [2, 4]]]
expected_mode = [[0, 1], [1, 2]]
input_ = tf.convert_to_tensor(sample, dtype=np.int32)
input_ph = tf.placeholder_with_default(
input=input_, shape=input_.shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph, event_ndims=2)
self.assertAllClose(self.evaluate(dist.mode()), expected_mode)
def testVarianceAndStd(self):
sample = [[[0, 1], [1, 2]], [[0, 3], [3, 6]]]
expected_variance = [[0, 1], [1, 4]]
input_ = tf.convert_to_tensor(sample, dtype=np.float64)
input_ph = tf.placeholder_with_default(
input=input_, shape=input_.shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph, event_ndims=2)
self.assertAllClose(self.evaluate(dist.variance()), expected_variance)
self.assertAllClose(self.evaluate(dist.stddev()),
np.sqrt(expected_variance))
def testMean(self):
samples = [[[0, 0, 1, 2], [0, 1, 1, 2]],
[[[0, 0], [1, 2]], [[0, 1], [2, 4]]]]
expected_means = [[0, 0.5, 1, 2], [[0.5, 1], [1, 2.5]]]
for sample, expected_mean in zip(samples, expected_means):
input_ = tf.convert_to_tensor(sample, dtype=np.float32)
input_ph = tf.placeholder_with_default(
input=input_,
shape=input_.shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph, event_ndims=1)
self.assertAllClose(self.evaluate(dist.mean()), expected_mean)
def testCdfWithBatch(self):
sample = [[0, 0, 1, 2], [0, 10, 20, 40]]
events = [[0], [0, 10], [[0, 1], [10, 20]]]
expected_cdfs = [[0.5, 0.25], [0.5, 0.5], [[0.5, 0.25], [1, 0.75]]]
for event, expected_cdf in zip(events, expected_cdfs):
input_ = tf.convert_to_tensor(sample, dtype=np.float32)
input_ph = tf.placeholder_with_default(
input=input_,
shape=input_.shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph)
self.assertAllClose(self.evaluate(dist.cdf(event)), expected_cdf)
self.assertAllClose(self.evaluate(dist.log_cdf(event)),
np.log(expected_cdf))
def testEntropy(self):
samples = [
[[0, 0], [0, 1]],
[[[0, 0], [0, 1], [0, 1], [1, 1]],
[[1, 1], [2, 2], [2, 2], [2, 2]]]
]
expected_entropys = [
entropy([0.5, 0.5]),
[entropy([0.25, 0.5, 0.25]), entropy([0.25, 0.75])]
]
for sample, expected_entropy in zip(samples, expected_entropys):
input_ = tf.convert_to_tensor(value=sample, dtype=np.float64)
input_ph = tf.compat.v1.placeholder_with_default(
input=input_, shape=input_.shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph, event_ndims=1)
self.assertAllClose(self.evaluate(dist.entropy()), expected_entropy)
def testSampleN(self):
for samples_shape in ([2], [2, 4], [4, 2, 4]):
input_ = random_samples(samples_shape)
input_ph = tf.compat.v1.placeholder_with_default(
input=input_, shape=samples_shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph)
self.assertAllEqual(
self.evaluate(tf.shape(input=dist.sample())),
dist.batch_shape_tensor())
n = 1000
seed = tf.compat.v1.set_random_seed(42) if tf.executing_eagerly() else 42
samples1 = dist.sample(n, seed)
seed = tf.compat.v1.set_random_seed(42) if tf.executing_eagerly() else 42
samples2 = dist.sample(n, seed)
self.assertAllEqual(
self.evaluate(samples1), self.evaluate(samples2))
def testVarianceAndStd(self):
samples = [[0, 1, 1, 2], [[1, 3], [2, 6]]]
expected_variances = [
0.5,
[1., 4.],
]
for sample, expected_variance in zip(samples, expected_variances):
input_ = tf.convert_to_tensor(sample, dtype=np.float32)
input_ph = tf.placeholder_with_default(
input=input_,
shape=input_.shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph)
self.assertAllClose(self.evaluate(dist.variance()),
expected_variance)
self.assertAllClose(self.evaluate(dist.stddev()),
np.sqrt(expected_variance))
def testMode(self):
samples = [
[0, 1, 1, 2],
[[0, 0, 1], [0, 1, 1], [2, 2, 2]],
[[[0, 0, 0, 0], [0, 0, 1, 1]],
[[1, 1, 1, 2], [0, 1, 2, 2]]]
]
expected_modes = [
1,
[0, 1, 2],
[[0, 0], [1, 2]]
]
for sample, expected_mode in zip(samples, expected_modes):
input_ = tf.convert_to_tensor(value=sample, dtype=np.int32)
input_ph = tf.compat.v1.placeholder_with_default(
input=input_, shape=input_.shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph)
self.assertAllClose(self.evaluate(dist.mode()), expected_mode)
def testSampleN(self):
for samples_shape in ([2, 2, 4], [4, 2, 2, 4]):
input_ = random_samples(samples_shape)
input_ph = tf.placeholder_with_default(
input=input_,
shape=samples_shape if self.static_shape else None)
dist = empirical.Empirical(samples=input_ph, event_ndims=2)
expected_shape = tf.concat(
[dist.batch_shape_tensor(),
dist.event_shape_tensor()], axis=0)
self.assertAllEqual(self.evaluate(tf.shape(dist.sample())),
expected_shape)
n = 1000
seed = tf.set_random_seed(42) if tf.executing_eagerly() else 42
samples1 = dist.sample(n, seed)
seed = tf.set_random_seed(42) if tf.executing_eagerly() else 42
samples2 = dist.sample(n, seed)
self.assertAllEqual(self.evaluate(samples1),
self.evaluate(samples2))
