def testDistribution(self, initial_known):
classes = np.random.randint(5, size=(20000, )) # Uniformly sampled
target_dist = [0.9, 0.05, 0.05, 0.0, 0.0]
initial_dist = [0.2] * 5 if initial_known else None
classes = math_ops.cast(classes,
dtypes.int64) # needed for Windows build.
dataset = dataset_ops.Dataset.from_tensor_slices(classes).shuffle(
200, seed=21).map(lambda c: (c, string_ops.as_string(c))).repeat()
get_next = self.getNext(
dataset.apply(
resampling.rejection_resample(target_dist=target_dist,
initial_dist=initial_dist,
class_func=lambda c, _: c,
seed=27)))
returned = []
while len(returned) < 4000:
returned.append(self.evaluate(get_next()))
returned_classes, returned_classes_and_data = zip(*returned)
_, returned_data = zip(*returned_classes_and_data)
self.assertAllEqual(
[compat.as_bytes(str(c)) for c in returned_classes], returned_data)
total_returned = len(returned_classes)
class_counts = np.array([
len([True for v in returned_classes if v == c]) for c in range(5)
])
returned_dist = class_counts / total_returned
self.assertAllClose(target_dist, returned_dist, atol=1e-2)
def benchmark_resample_performance(self):
init_dist = [0.25, 0.25, 0.25, 0.25]
target_dist = [0.0, 0.0, 0.0, 1.0]
num_classes = len(init_dist)
# We don't need many samples to test a dirac-delta target distribution
num_samples = 1000
data_np = np.random.choice(num_classes, num_samples, p=init_dist)
# Prepare the dataset
dataset = dataset_ops.Dataset.from_tensor_slices(data_np).repeat()
# Reshape distribution via rejection sampling.
dataset = dataset.apply(
resampling.rejection_resample(class_func=lambda x: x,
target_dist=target_dist,
initial_dist=init_dist,
seed=142))
options = dataset_ops.Options()
options.experimental_optimization.apply_default_optimizations = False
dataset = dataset.with_options(options)
wall_time = self.run_benchmark(dataset=dataset,
num_elements=num_samples,
iters=10,
warmup=True)
resample_time = wall_time * num_samples
self.report_benchmark(iters=10,
wall_time=resample_time,
name="resample_{}".format(num_samples))
def testDistribution(self, initial_known):
classes = np.random.randint(5, size=(20000,)) # Uniformly sampled
target_dist = [0.9, 0.05, 0.05, 0.0, 0.0]
initial_dist = [0.2] * 5 if initial_known else None
classes = math_ops.cast(classes, dtypes.int64) # needed for Windows build.
dataset = dataset_ops.Dataset.from_tensor_slices(classes).shuffle(
200, seed=21).map(lambda c: (c, string_ops.as_string(c))).repeat()
get_next = self.getNext(
dataset.apply(
resampling.rejection_resample(
target_dist=target_dist,
initial_dist=initial_dist,
class_func=lambda c, _: c,
seed=27)))
returned = []
while len(returned) < 4000:
returned.append(self.evaluate(get_next()))
returned_classes, returned_classes_and_data = zip(*returned)
_, returned_data = zip(*returned_classes_and_data)
self.assertAllEqual([compat.as_bytes(str(c))
for c in returned_classes], returned_data)
total_returned = len(returned_classes)
class_counts = np.array([
len([True for v in returned_classes if v == c])
for c in range(5)])
returned_dist = class_counts / total_returned
self.assertAllClose(target_dist, returned_dist, atol=1e-2)
def testOtherDtypes(self, target_dtype, init_dtype):
target_dist = np.array([0.5, 0.5], dtype=target_dtype)
if init_dtype is None:
init_dist = None
else:
init_dist = np.array([0.5, 0.5], dtype=init_dtype)
dataset = dataset_ops.Dataset.range(10)
resampler = resampling.rejection_resample(class_func=lambda x: x % 2,
target_dist=target_dist,
initial_dist=init_dist)
dataset = dataset.apply(resampler)
get_next = self.getNext(dataset)
self.evaluate(get_next())
def _time_resampling(test_obj, data_np, target_dist, init_dist, num_to_sample):
dataset = dataset_ops.Dataset.from_tensor_slices(data_np).repeat()
# Reshape distribution via rejection sampling.
dataset = dataset.apply(
resampling.rejection_resample(class_func=lambda x: x,
target_dist=target_dist,
initial_dist=init_dist,
seed=142))
get_next = dataset_ops.make_one_shot_iterator(dataset).get_next()
with test_obj.test_session() as sess:
start_time = time.time()
for _ in xrange(num_to_sample):
sess.run(get_next)
end_time = time.time()
return end_time - start_time
def rejection_resample(class_func, target_dist, initial_dist=None, seed=None):
"""A transformation that resamples a dataset to achieve a target distribution.
**NOTE** Resampling is performed via rejection sampling; some fraction
of the input values will be dropped.
Args:
class_func: A function mapping an element of the input dataset to a scalar
`tf.int32` tensor. Values should be in `[0, num_classes)`.
target_dist: A floating point type tensor, shaped `[num_classes]`.
initial_dist: (Optional.) A floating point type tensor, shaped
`[num_classes]`. If not provided, the true class distribution is
estimated live in a streaming fashion.
seed: (Optional.) Python integer seed for the resampler.
Returns:
A `Dataset` transformation function, which can be passed to
`tf.data.Dataset.apply`.
"""
return resampling.rejection_resample(class_func, target_dist, initial_dist,
seed)
def _time_resampling(
test_obj, data_np, target_dist, init_dist, num_to_sample):
dataset = dataset_ops.Dataset.from_tensor_slices(data_np).repeat()
# Reshape distribution via rejection sampling.
dataset = dataset.apply(
resampling.rejection_resample(
class_func=lambda x: x,
target_dist=target_dist,
initial_dist=init_dist,
seed=142))
get_next = dataset.make_one_shot_iterator().get_next()
with test_obj.test_session() as sess:
start_time = time.time()
for _ in xrange(num_to_sample):
sess.run(get_next)
end_time = time.time()
return end_time - start_time
def rejection_resample(class_func, target_dist, initial_dist=None, seed=None):
"""A transformation that resamples a dataset to achieve a target distribution.
**NOTE** Resampling is performed via rejection sampling; some fraction
of the input values will be dropped.
Args:
class_func: A function mapping an element of the input dataset to a scalar
`tf.int32` tensor. Values should be in `[0, num_classes)`.
target_dist: A floating point type tensor, shaped `[num_classes]`.
initial_dist: (Optional.) A floating point type tensor, shaped
`[num_classes]`. If not provided, the true class distribution is
estimated live in a streaming fashion.
seed: (Optional.) Python integer seed for the resampler.
Returns:
A `Dataset` transformation function, which can be passed to
`tf.data.Dataset.apply`.
"""
return resampling.rejection_resample(class_func, target_dist, initial_dist,
seed)
def testExhaustion(self):
init_dist = [0.5, 0.5]
target_dist = [0.9, 0.1]
dataset = dataset_ops.Dataset.range(10000)
resampler = resampling.rejection_resample(class_func=lambda x: x % 2,
target_dist=target_dist,
initial_dist=init_dist)
dataset = dataset.apply(resampler)
get_next = self.getNext(dataset)
returned = []
with self.assertRaises(errors.OutOfRangeError):
while True:
returned.append(self.evaluate(get_next()))
classes, _ = zip(*returned)
bincount = np.bincount(np.array(classes),
minlength=len(init_dist)).astype(
np.float32) / len(classes)
self.assertAllClose(target_dist, bincount, atol=1e-2)
def testEdgeCasesSampleFromInitialDataset(self, only_initial_dist):
init_dist = [0.5, 0.5]
target_dist = [0.5, 0.5] if only_initial_dist else [0.0, 1.0]
num_classes = len(init_dist)
# We don't need many samples to test that this works.
num_samples = 100
data_np = np.random.choice(num_classes, num_samples, p=init_dist)
dataset = dataset_ops.Dataset.from_tensor_slices(data_np)
# Reshape distribution.
dataset = dataset.apply(
resampling.rejection_resample(class_func=lambda x: x,
target_dist=target_dist,
initial_dist=init_dist))
get_next = self.getNext(dataset)
returned = []
with self.assertRaises(errors.OutOfRangeError):
while True:
returned.append(self.evaluate(get_next()))
def _time_resampling(data_np, target_dist, init_dist, num_to_sample): # pylint: disable=missing-docstring
dataset = dataset_ops.Dataset.from_tensor_slices(data_np).repeat()
# Reshape distribution via rejection sampling.
dataset = dataset.apply(
resampling.rejection_resample(class_func=lambda x: x,
target_dist=target_dist,
initial_dist=init_dist,
seed=142))
options = dataset_ops.Options()
options.experimental_optimization.apply_default_optimizations = False
dataset = dataset.with_options(options)
get_next = dataset_ops.make_one_shot_iterator(dataset).get_next()
with session.Session() as sess:
start_time = time.time()
for _ in xrange(num_to_sample):
sess.run(get_next)
end_time = time.time()
return end_time - start_time
def testEdgeCasesSampleFromInitialDataset(self, only_initial_dist):
init_dist = [0.5, 0.5]
target_dist = [0.5, 0.5] if only_initial_dist else [0.0, 1.0]
num_classes = len(init_dist)
# We don't need many samples to test that this works.
num_samples = 100
data_np = np.random.choice(num_classes, num_samples, p=init_dist)
dataset = dataset_ops.Dataset.from_tensor_slices(data_np)
# Reshape distribution.
dataset = dataset.apply(
resampling.rejection_resample(
class_func=lambda x: x,
target_dist=target_dist,
initial_dist=init_dist))
get_next = self.getNext(dataset)
returned = []
with self.assertRaises(errors.OutOfRangeError):
while True:
returned.append(self.evaluate(get_next()))
def _time_resampling(data_np, target_dist, init_dist, num_to_sample): # pylint: disable=missing-docstring
dataset = dataset_ops.Dataset.from_tensor_slices(data_np).repeat()
# Reshape distribution via rejection sampling.
dataset = dataset.apply(
resampling.rejection_resample(
class_func=lambda x: x,
target_dist=target_dist,
initial_dist=init_dist,
seed=142))
options = dataset_ops.Options()
options.experimental_optimization.apply_default_optimizations = False
dataset = dataset.with_options(options)
get_next = dataset_ops.make_one_shot_iterator(dataset).get_next()
with session.Session() as sess:
start_time = time.time()
for _ in xrange(num_to_sample):
sess.run(get_next)
end_time = time.time()
return end_time - start_time
def testRandomClasses(self):
init_dist = [0.25, 0.25, 0.25, 0.25]
target_dist = [0.0, 0.0, 0.0, 1.0]
num_classes = len(init_dist)
# We don't need many samples to test a dirac-delta target distribution.
num_samples = 100
data_np = np.random.choice(num_classes, num_samples, p=init_dist)
dataset = dataset_ops.Dataset.from_tensor_slices(data_np)
# Apply a random mapping that preserves the data distribution.
def _remap_fn(_):
return math_ops.cast(
random_ops.random_uniform([1]) * num_classes, dtypes.int32)[0]
dataset = dataset.map(_remap_fn)
# Reshape distribution.
dataset = dataset.apply(
resampling.rejection_resample(class_func=lambda x: x,
target_dist=target_dist,
initial_dist=init_dist))
get_next = dataset_ops.make_one_shot_iterator(dataset).get_next()
with self.cached_session() as sess:
returned = []
with self.assertRaises(errors.OutOfRangeError):
while True:
returned.append(sess.run(get_next))
classes, _ = zip(*returned)
bincount = np.bincount(np.array(classes),
minlength=num_classes).astype(
np.float32) / len(classes)
self.assertAllClose(target_dist, bincount, atol=1e-2)
def testRandomClasses(self):
init_dist = [0.25, 0.25, 0.25, 0.25]
target_dist = [0.0, 0.0, 0.0, 1.0]
num_classes = len(init_dist)
# We don't need many samples to test a dirac-delta target distribution.
num_samples = 100
data_np = np.random.choice(num_classes, num_samples, p=init_dist)
dataset = dataset_ops.Dataset.from_tensor_slices(data_np)
# Apply a random mapping that preserves the data distribution.
def _remap_fn(_):
return math_ops.cast(random_ops.random_uniform([1]) * num_classes,
dtypes.int32)[0]
dataset = dataset.map(_remap_fn)
# Reshape distribution.
dataset = dataset.apply(
resampling.rejection_resample(
class_func=lambda x: x,
target_dist=target_dist,
initial_dist=init_dist))
get_next = dataset.make_one_shot_iterator().get_next()
with self.cached_session() as sess:
returned = []
with self.assertRaises(errors.OutOfRangeError):
while True:
returned.append(sess.run(get_next))
classes, _ = zip(*returned)
bincount = np.bincount(
np.array(classes),
minlength=num_classes).astype(np.float32) / len(classes)
self.assertAllClose(target_dist, bincount, atol=1e-2)
