def testOptimizationNonSerializable(self): dataset = dataset_ops.Dataset.from_tensors(0) dataset = dataset.apply(optimization.assert_next(["FiniteSkip"])) dataset = dataset.skip(0) # Should not be removed by noop elimination dataset = dataset.apply(optimization.non_serializable()) dataset = dataset.apply(optimization.assert_next(["MemoryCacheImpl"])) dataset = dataset.skip(0) # Should be removed by noop elimination dataset = dataset.cache() dataset = dataset_ops._OptimizeDataset(dataset, ["noop_elimination"]) self.assertDatasetProduces(dataset, expected_output=[0])
def testOptimizationNonSerializable(self): dataset = dataset_ops.Dataset.from_tensors(0) dataset = dataset.apply(optimization.assert_next(["FiniteSkip"])) dataset = dataset.skip(0) # Should not be removed by noop elimination dataset = dataset.apply(optimization.non_serializable()) dataset = dataset.apply(optimization.assert_next(["MemoryCacheImpl"])) dataset = dataset.skip(0) # Should be removed by noop elimination dataset = dataset.cache() options = dataset_ops.Options() options.experimental_optimization.apply_default_optimizations = False options.experimental_optimization.noop_elimination = True dataset = dataset.with_options(options) self.assertDatasetProduces(dataset, expected_output=[0])
def testFilterFusion(self, map_function, predicates):
dataset = dataset_ops.Dataset.range(5).apply(
optimization.assert_next(["Map", "Filter",
"MemoryCacheImpl"])).map(map_function)
for predicate in predicates:
dataset = dataset.filter(predicate)
dataset = dataset.cache()
options = dataset_ops.Options()
options.experimental_optimization.apply_default_optimizations = False
options.experimental_optimization.filter_fusion = True
dataset = dataset.with_options(options)
expected_output = []
for x in range(5):
r = map_function(x)
filtered = False
for predicate in predicates:
if isinstance(r, tuple):
b = predicate(*r) # Pass tuple as multiple arguments.
else:
b = predicate(r)
if not self.evaluate(b):
filtered = True
break
if not filtered:
expected_output.append(r)
self.assertDatasetProduces(dataset, expected_output=expected_output)
def flat_map_fn(_): dataset = dataset_ops.Dataset.from_tensors(0) dataset = dataset.apply(optimization.assert_next(["MapAndBatch"])) # Should be fused by map and batch fusion dataset = dataset.map(lambda x: x) dataset = dataset.batch(1) return dataset
def testNoopElimination(self):
a = constant_op.constant(1, dtype=dtypes.int64)
b = constant_op.constant(2, dtype=dtypes.int64)
some_tensor = math_ops.mul(a, b)
dataset = dataset_ops.Dataset.range(5)
dataset = dataset.apply(
optimization.assert_next(
["FiniteRepeat", "FiniteSkip", "Prefetch", "Prefetch"]))
dataset = dataset.repeat(some_tensor).skip(5).prefetch(0).take(-1).skip(
0).repeat(1).prefetch(0)
options = dataset_ops.Options()
options.experimental_noop_elimination = True
dataset = dataset.with_options(options)
iterator = dataset.make_one_shot_iterator()
get_next = iterator.get_next()
with self.test_session() as sess:
for x in range(5):
result = sess.run(get_next)
self.assertAllEqual(result, x)
with self.assertRaises(errors.OutOfRangeError):
sess.run(get_next)
def testFilterFusion(self, map_function, predicates):
dataset = dataset_ops.Dataset.range(5).apply(
optimization.assert_next(["Map", "Filter",
"MemoryCacheImpl"])).map(map_function)
for predicate in predicates:
dataset = dataset.filter(predicate)
dataset = dataset.cache()
options = dataset_ops.Options()
options.experimental_filter_fusion = True
dataset = dataset.with_options(options)
iterator = dataset.make_one_shot_iterator()
get_next = iterator.get_next()
with self.cached_session() as sess:
for x in range(5):
r = map_function(x)
filtered = False
for predicate in predicates:
if isinstance(r, tuple):
b = predicate(*r) # Pass tuple as multiple arguments.
else:
b = predicate(r)
if not sess.run(b):
filtered = True
break
if not filtered:
result = sess.run(get_next)
self.assertAllEqual(r, result)
with self.assertRaises(errors.OutOfRangeError):
sess.run(get_next)
def testMapFusion(self, functions):
dataset = dataset_ops.Dataset.range(5).apply(
optimization.assert_next(["Map", "MemoryCacheImpl"]))
for function in functions:
dataset = dataset.map(function)
dataset = dataset.cache()
options = dataset_ops.Options()
options.experimental_map_fusion = True
dataset = dataset.with_options(options)
iterator = dataset.make_one_shot_iterator()
get_next = iterator.get_next()
with self.cached_session() as sess:
for x in range(5):
result = sess.run(get_next)
r = x
for function in functions:
if isinstance(r, tuple):
r = function(*r) # Pass tuple as multiple arguments.
else:
r = function(r)
self.assertAllEqual(r, result)
with self.assertRaises(errors.OutOfRangeError):
sess.run(get_next)
def testAssertNext(self):
dataset = dataset_ops.Dataset.from_tensors(0).apply(
optimization.assert_next(["Map"])).map(lambda x: x)
options = dataset_ops.Options()
options.experimental_optimization.apply_default_optimizations = False
dataset = dataset.with_options(options)
self.assertDatasetProduces(dataset, expected_output=[0])
def testLatencyStatsOptimization(self):
stats_aggregator = stats_ops.StatsAggregator()
dataset = dataset_ops.Dataset.from_tensors(1).apply(
optimization.assert_next(
["LatencyStats", "Map", "LatencyStats", "Prefetch",
"LatencyStats"])).map(lambda x: x * x).prefetch(1).apply(
stats_ops.set_stats_aggregator(stats_aggregator))
options = dataset_ops.Options()
options.experimental_latency_all_edges = True
dataset = dataset.with_options(options)
iterator = dataset.make_initializable_iterator()
get_next = iterator.get_next()
summary_t = stats_aggregator.get_summary()
with self.cached_session() as sess:
sess.run(iterator.initializer)
self.assertEqual(1 * 1, sess.run(get_next))
with self.assertRaises(errors.OutOfRangeError):
sess.run(get_next)
summary_str = sess.run(summary_t)
self._assertSummaryHasCount(summary_str,
"record_latency_TensorDataset/_1", 1)
self._assertSummaryHasCount(summary_str, "record_latency_MapDataset/_4",
1)
self._assertSummaryHasCount(summary_str,
"record_latency_PrefetchDataset/_6", 1)
def testMapFilterFusion(self, function, predicate):
dataset = dataset_ops.Dataset.range(10).apply(
optimization.assert_next(
["Map", "FilterByLastComponent"])).map(function).filter(predicate)
options = dataset_ops.Options()
options.experimental_map_and_filter_fusion = True
dataset = dataset.with_options(options)
self._testMapAndFilter(dataset, function, predicate)
def testAssertNext(self):
dataset = dataset_ops.Dataset.from_tensors(0).apply(
optimization.assert_next(["Map"])).map(lambda x: x)
iterator = dataset.make_one_shot_iterator()
get_next = iterator.get_next()
with self.cached_session() as sess:
self.assertEqual(0, sess.run(get_next))
def testAssertNextInvalid(self):
dataset = dataset_ops.Dataset.from_tensors(0).apply(
optimization.assert_next(["Whoops"])).map(lambda x: x)
self.assertDatasetProduces(
dataset,
expected_error=(
errors.InvalidArgumentError,
"Asserted Whoops transformation at offset 0 but encountered "
"Map transformation instead."))
def testMakeNumaAware(self):
dataset = dataset_ops.Dataset.range(10).apply(
optimization.assert_next(["NumaMapAndBatch"])).apply(
batching.map_and_batch(lambda x: x * x, 10))
options = dataset_ops.Options()
options.experimental_numa_aware = True
dataset = dataset.with_options(options)
self.assertDatasetProduces(
dataset, expected_output=[[x * x for x in range(10)]])
def testMapAndBatchFusion(self):
dataset = dataset_ops.Dataset.range(10).apply(
optimization.assert_next(
["MapAndBatch"])).map(lambda x: x * x).batch(10)
options = dataset_ops.Options()
options.experimental_map_and_batch_fusion = True
dataset = dataset.with_options(options)
self.assertDatasetProduces(
dataset, expected_output=[[x * x for x in range(10)]])
def testHoisting(self, function, will_optimize):
dataset = dataset_ops.Dataset.range(5).apply(
optimization.assert_next(
["Zip[0]", "Map"] if will_optimize else ["Map"])).map(function)
options = dataset_ops.Options()
options.experimental_hoist_random_uniform = True
dataset = dataset.with_options(options)
self._testDataset(dataset)
def _make_dataset(node_names): dataset = base_dataset.apply(optimization.assert_next(node_names)) dataset = dataset.map(map_fn, num_parallel_calls) dataset = dataset.batch(100) options = dataset_ops.Options() options.experimental_optimization.apply_default_optimizations = False options.experimental_optimization.map_and_batch_fusion = False dataset = dataset.with_options(options) return dataset
def testMapParallelization(self, function, should_optimize):
next_nodes = ["ParallelMap"] if should_optimize else ["Map"]
dataset = dataset_ops.Dataset.range(5).apply(
optimization.assert_next(next_nodes)).map(function)
options = dataset_ops.Options()
options.experimental_map_parallelization = True
dataset = dataset.with_options(options)
if should_optimize:
self.assertDatasetProduces(
dataset, expected_output=[function(x) for x in range(5)])
def testAssertNextShort(self):
dataset = dataset_ops.Dataset.from_tensors(0).apply(
optimization.assert_next(["Map", "Whoops"])).map(lambda x: x)
iterator = dataset.make_one_shot_iterator()
get_next = iterator.get_next()
with self.cached_session() as sess:
with self.assertRaisesRegexp(
errors.InvalidArgumentError,
"Asserted next 2 transformations but encountered only 1."):
sess.run(get_next)
def testAssertNextShort(self):
dataset = dataset_ops.Dataset.from_tensors(0).apply(
optimization.assert_next(["Map", "Whoops"])).map(lambda x: x)
options = dataset_ops.Options()
options.experimental_autotune = False
dataset = dataset.with_options(options)
self.assertDatasetProduces(
dataset,
expected_error=(
errors.InvalidArgumentError,
"Asserted next 2 transformations but encountered only 1."))
def testOptimizationDefault(self):
dataset = dataset_ops.Dataset.range(10).apply(
optimization.assert_next(["Map",
"Batch"])).map(lambda x: x * x).batch(10)
iterator = dataset.with_options(
dataset_ops.Options()).make_one_shot_iterator()
get_next = iterator.get_next()
with self.cached_session() as sess:
self.assertAllEqual([x * x for x in range(10)], sess.run(get_next))
with self.assertRaises(errors.OutOfRangeError):
sess.run(get_next)
def testFilterWithRandomUniformFusion(self):
dataset = dataset_ops.Dataset.range(10000000).apply(
optimization.assert_next(["Sampling"]))
dataset = dataset.filter(lambda _: random_ops.random_uniform([]) < 0.05)
options = dataset_ops.Options()
options.experimental_optimization.apply_default_optimizations = False
options.experimental_optimization.filter_with_random_uniform_fusion = True
dataset = dataset.with_options(options)
get_next = self.getNext(dataset)
self.evaluate(get_next())
def testAutotuneOption(self):
dataset = dataset_ops.Dataset.from_tensors(0)
dataset = dataset.map(lambda x: x).apply(
optimization.assert_next(["Model"]))
options = dataset_ops.Options()
options.experimental_autotune = True
options.experimental_optimization.apply_default_optimizations = False
dataset = dataset.with_options(options)
get_next = self.getNext(dataset)
self.assertEqual(0, self.evaluate(get_next()))
with self.assertRaises(errors.OutOfRangeError):
self.evaluate(get_next())
def testMapParallelizationWithCapturedConstant(self):
"""Tests that functions with captured constants are parallelized."""
captured_t = constant_op.constant(42, dtype=dtypes.int64)
def fn(x):
return x + captured_t
dataset = dataset_ops.Dataset.range(5).apply(
optimization.assert_next(["ParallelMap"])).map(fn)
options = dataset_ops.Options()
options.experimental_optimization.apply_default_optimizations = False
options.experimental_optimization.map_parallelization = True
dataset = dataset.with_options(options)
self.assertDatasetProduces(
dataset, expected_output=[x + 42 for x in range(5)])
def testMapAndBatchFusion(self):
dataset = dataset_ops.Dataset.range(10).apply(
optimization.assert_next(
["MapAndBatch"])).map(lambda x: x * x).batch(10)
options = dataset_ops.Options()
options.experimental_map_and_batch_fusion = True
dataset = dataset.with_options(options)
iterator = dataset.make_one_shot_iterator()
get_next = iterator.get_next()
with self.cached_session() as sess:
self.assertAllEqual([x * x for x in range(10)], sess.run(get_next))
with self.assertRaises(errors.OutOfRangeError):
sess.run(get_next)
def testNoopElimination(self):
a = constant_op.constant(1, dtype=dtypes.int64)
b = constant_op.constant(2, dtype=dtypes.int64)
some_tensor = math_ops.mul(a, b)
dataset = dataset_ops.Dataset.range(5)
dataset = dataset.apply(
optimization.assert_next(
["FiniteRepeat", "FiniteSkip", "Prefetch", "MemoryCacheImpl"]))
dataset = dataset.repeat(some_tensor).skip(5).take(-1).skip(0).repeat(
1).prefetch(0).prefetch(1).cache()
options = dataset_ops.Options()
options.experimental_noop_elimination = True
dataset = dataset.with_options(options)
self.assertDatasetProduces(dataset, expected_output=range(5))
def testAutotuneOption(self):
dataset = dataset_ops.Dataset.from_tensors(0)
dataset = dataset.map(lambda x: x).apply(
optimization.assert_next(["Model"]))
options = dataset_ops.Options()
options.experimental_autotune = True
dataset = dataset.with_options(options)
iterator = dataset_ops.make_one_shot_iterator(dataset)
get_next = iterator.get_next()
with self.cached_session() as sess:
self.assertEqual(0, self.evaluate(get_next))
with self.assertRaises(errors.OutOfRangeError):
self.evaluate(get_next)
def testAdditionalInputs(self):
a = constant_op.constant(1, dtype=dtypes.float32)
b = constant_op.constant(0, dtype=dtypes.float32)
some_tensor = math_ops.mul(a, b)
def random_with_capture(_):
return some_tensor + random_ops.random_uniform(
[], minval=1, maxval=10, dtype=dtypes.float32, seed=42)
dataset = dataset_ops.Dataset.range(5).apply(
optimization.assert_next(["Zip[0]", "Map"])).map(random_with_capture)
options = dataset_ops.Options()
options.experimental_hoist_random_uniform = True
dataset = dataset.with_options(options)
self._testDataset(dataset)
def testMapParallelizationWithCapturedVariable(self):
"""Tests that functions with captured variables are not parallelized."""
captured_t = variables.Variable(42, dtype=dtypes.int64)
def fn(x):
return x + captured_t
dataset = dataset_ops.Dataset.range(5).apply(
optimization.assert_next(["Map"])).map(fn)
options = dataset_ops.Options()
options.experimental_optimization.apply_default_optimizations = False
options.experimental_optimization.map_parallelization = True
dataset = dataset.with_options(options)
self.evaluate(variables.global_variables_initializer())
self.assertDatasetProduces(
dataset,
expected_output=[x + 42 for x in range(5)],
requires_initialization=True)
def testCapturedInputs(self):
a = constant_op.constant(3, dtype=dtypes.int64)
b = constant_op.constant(4, dtype=dtypes.int64)
some_tensor = math_ops.mul(a, b)
function = lambda x: x * x
def predicate(y):
return math_ops.less(math_ops.cast(y, dtypes.int64), some_tensor)
# We are currently not supporting functions with captured inputs.
dataset = dataset_ops.Dataset.range(10).apply(
optimization.assert_next(["Map",
"Filter"])).map(function).filter(predicate)
options = dataset_ops.Options()
options.experimental_map_and_filter_fusion = True
dataset = dataset.with_options(options)
self._testMapAndFilter(dataset, function, predicate)
def testShuffleAndRepeatFusion(self):
dataset = dataset_ops.Dataset.range(10).apply(
optimization.assert_next(["ShuffleAndRepeat"])).shuffle(10).repeat(2)
options = dataset_ops.Options()
options.experimental_shuffle_and_repeat_fusion = True
dataset = dataset.with_options(options)
get_next = self.getNext(dataset)
for _ in range(2):
results = []
for _ in range(10):
results.append(self.evaluate(get_next()))
self.assertAllEqual([x for x in range(10)], sorted(results))
with self.assertRaises(errors.OutOfRangeError):
self.evaluate(get_next())
with self.assertRaises(errors.OutOfRangeError):
self.evaluate(get_next())
def testOptimization(self):
dataset = dataset_ops.Dataset.range(10)
dataset = dataset.apply(optimization.assert_next(["MemoryCacheImpl"]))
dataset = dataset.skip(0) # this should be optimized away
dataset = dataset.cache()
options = dataset_ops.Options()
options.experimental_optimization.noop_elimination = True
dataset = dataset.with_options(options)
multi_device_iterator = multi_device_iterator_ops.MultiDeviceIterator(
dataset, ["/cpu:1", "/cpu:2"])
config = config_pb2.ConfigProto(device_count={"CPU": 3})
with self.test_session(config=config):
self.evaluate(multi_device_iterator.initializer)
for i in range(0, 10, 2):
elem_on_1, elem_on_2 = multi_device_iterator.get_next()
self.assertEqual(i, self.evaluate(elem_on_1))
self.assertEqual(i + 1, self.evaluate(elem_on_2))
with self.assertRaises(errors.OutOfRangeError):
elem_on_1, elem_on_2 = multi_device_iterator.get_next()
self.evaluate(elem_on_1)
self.evaluate(elem_on_2)
def testNoopElimination(self):
a = constant_op.constant(1, dtype=dtypes.int64)
b = constant_op.constant(2, dtype=dtypes.int64)
some_tensor = math_ops.mul(a, b)
dataset = dataset_ops.Dataset.range(5)
dataset = dataset.apply(
optimization.assert_next(
["FiniteRepeat", "FiniteSkip", "Prefetch", "MemoryCacheImpl"]))
dataset = dataset.repeat(some_tensor).skip(5).take(-1).skip(0).repeat(
1).prefetch(0).prefetch(1).cache()
options = dataset_ops.Options()
options.experimental_noop_elimination = True
dataset = dataset.with_options(options)
iterator = dataset.make_one_shot_iterator()
get_next = iterator.get_next()
with self.test_session() as sess:
for x in range(5):
result = sess.run(get_next)
self.assertAllEqual(result, x)
with self.assertRaises(errors.OutOfRangeError):
sess.run(get_next)
def testLatencyStatsOptimization(self):
aggregator = stats_aggregator.StatsAggregator()
dataset = dataset_ops.Dataset.from_tensors(1).apply(
optimization.assert_next([
"LatencyStats", "Map", "LatencyStats", "Prefetch",
"LatencyStats"
])).map(lambda x: x * x).prefetch(1)
options = dataset_ops.Options()
options.experimental_optimization.apply_default_optimizations = False
options.experimental_stats.latency_all_edges = True
options.experimental_stats.aggregator = aggregator
dataset = dataset.with_options(options)
self.assertDatasetProduces(dataset,
expected_output=[1],
requires_initialization=True,
num_test_iterations=1)
handle = self.getHandle(aggregator)
self.assertStatisticsHasCount(
handle, self.regexForNodeName("record_latency::TensorDataset"), 1)
self.assertStatisticsHasCount(
handle, self.regexForNodeName("record_latency::MapDataset"), 1)
self.assertStatisticsHasCount(
handle, self.regexForNodeName("record_latency::PrefetchDataset"),
1)
def testLatencyStatsOptimization(self):
aggregator = stats_aggregator.StatsAggregator()
dataset = dataset_ops.Dataset.from_tensors(1).apply(
optimization.assert_next([
"LatencyStats", "Map", "LatencyStats", "Prefetch",
"LatencyStats"
])).map(lambda x: x * x).prefetch(1)
options = dataset_ops.Options()
options.experimental_stats = stats_options.StatsOptions()
options.experimental_stats.latency_all_edges = True
options.experimental_stats.aggregator = aggregator
dataset = dataset.with_options(options)
self.assertDatasetProduces(dataset,
expected_output=[1],
requires_initialization=True,
num_test_iterations=1)
summary_t = aggregator.get_summary()
summary_str = self.evaluate(summary_t)
self._assertSummaryHasCount(summary_str,
"record_latency_TensorDataset/_1", 1)
self._assertSummaryHasCount(summary_str,
"record_latency_MapDataset/_4", 1)
self._assertSummaryHasCount(summary_str,
"record_latency_PrefetchDataset/_6", 1)
def make_dataset(node_names):
dataset = base_dataset.apply(optimization.assert_next(node_names))
dataset = apply_fn_1(dataset)
dataset = apply_fn_2(dataset)
return dataset
def _make_dataset(node_names):
dataset = base_dataset.apply(optimization.assert_next(node_names))
dataset = dataset.apply(batching.map_and_batch(map_fn, 100))
return dataset
def _make_dataset(node_names):
return base_dataset.apply(optimization.assert_next(node_names)).map(
map_fn, num_parallel_calls=num_parallel_calls).batch(batch_size)
def testNoRegularMap(self): dataset = dataset_ops.Dataset.range(100) dataset = dataset.apply(optimization.assert_next(["Map", "FiniteTake"])) dataset = dataset.map(lambda x: x + 1).take(50) dataset = self._enable_autotune_buffers(dataset) self.assertDatasetProduces(dataset, range(1, 51))
def testOptions(self):
dataset = dataset_ops.Dataset.range(5)
dataset = dataset.apply(optimization.assert_next(["MapAndBatch"]))
dataset = dataset.map(lambda x: x).batch(5)
self.evaluate(dataset.reduce(0, lambda state, value: state))
def flat_map_fn(_): dataset = dataset_ops.Dataset.from_tensors(0) dataset = dataset.apply(optimization.assert_next(["MemoryCacheImpl"])) dataset = dataset.skip(0) # Should be removed by noop elimination dataset = dataset.cache() return dataset
def testAssertNext(self):
dataset = dataset_ops.Dataset.from_tensors(0).apply(
optimization.assert_next(["Map"])).map(lambda x: x)
self.assertDatasetProduces(dataset, expected_output=[0])
