[GitHub] [beam] yeandy commented on a change in pull request #16970: [BEAM-13982] A base class for run inference

[GitBox <gi...@apache.org>]

yeandy commented on a change in pull request #16970:
URL: https://github.com/apache/beam/pull/16970#discussion_r817956018



##########
File path: sdks/python/apache_beam/ml/inference/base.py
##########
@@ -0,0 +1,194 @@
+import logging
+import platform
+import resource
+import sys
+import time
+from typing import Any
+from typing import List
+
+import apache_beam as beam
+from apache_beam.utils import shared
+from apache_beam.ml.inference.apis import PredictionResult
+
+_MILLISECOND_TO_MICROSECOND = 1000
+_MICROSECOND_TO_NANOSECOND = 1000
+_SECOND_TO_MICROSECOND = 1000000
+
+
+def _unbatch(maybe_keyed_batches: Any):
+  keys, results = maybe_keyed_batches
+  if keys:
+    return zip(keys, results)
+  else:
+    return results
+
+
+class ModelLoader:
+  """Has the ability to load an ML model."""
+  def load_model(self):
+    """Loads an initializes a model for processing."""
+    raise NotImplementedError(type(self))
+
+
+class InferenceRunner:
+  """Implements running inferences for a framework."""
+  def run_inference(self, batch: Any, model: Any) -> List[PredictionResult]:

Review comment:
       Do we want `Iterable[PredictionResult]` instead of `List[PredictionResult]`?

##########
File path: sdks/python/apache_beam/ml/inference/base.py
##########
@@ -0,0 +1,194 @@
+import logging
+import platform
+import resource
+import sys
+import time
+from typing import Any
+from typing import List
+
+import apache_beam as beam
+from apache_beam.utils import shared
+from apache_beam.ml.inference.apis import PredictionResult

Review comment:
       I have `api.py`. Do we we want with the `s`? 

##########
File path: sdks/python/apache_beam/ml/inference/base.py
##########
@@ -0,0 +1,194 @@
+import logging
+import platform
+import resource
+import sys
+import time
+from typing import Any
+from typing import List
+
+import apache_beam as beam
+from apache_beam.utils import shared
+from apache_beam.ml.inference.apis import PredictionResult
+
+_MILLISECOND_TO_MICROSECOND = 1000
+_MICROSECOND_TO_NANOSECOND = 1000
+_SECOND_TO_MICROSECOND = 1000000
+
+
+def _unbatch(maybe_keyed_batches: Any):
+  keys, results = maybe_keyed_batches
+  if keys:
+    return zip(keys, results)
+  else:
+    return results
+
+
+class ModelLoader:
+  """Has the ability to load an ML model."""
+  def load_model(self):
+    """Loads an initializes a model for processing."""
+    raise NotImplementedError(type(self))
+
+
+class InferenceRunner:

Review comment:
       Robert had something like 
   ```
   ExampleBatchType = TypeVar('ExampleBatchType')
   InferenceBatchType = TypeVar('InferenceBatchType')
    
   class InferenceRunner(Generic[ExampleBatchType, InferenceBatchType]):
     def to_impl_batch(batch: Iterable[Any] -> ExampleBatchType):
       raise NotImplementedError(type(self))
    
     def from_impl_batch(inferences: InferenceBatchType) -> Iterable[Any]:
       raise NotImplementedError(type(self))
    
     def run_inference(self, batch : ExampleBatchType) -> InferenceBatchType:
       raise NotImplementedError(type(self))
   ```
   I'm guessing it's not necessary to have the type variables if you're not converting to and from batches? But how do we structure this if we want to leverage Batched DoFn in the future? Thought it's probably not necessary to do the batching explicitly if we can leverage something like `@DoFn.yields_batches`? @TheNeuralBit what are your thoughts?

##########
File path: sdks/python/apache_beam/ml/inference/base.py
##########
@@ -0,0 +1,194 @@
+import logging
+import platform
+import resource
+import sys
+import time
+from typing import Any
+from typing import List
+
+import apache_beam as beam
+from apache_beam.utils import shared
+from apache_beam.ml.inference.apis import PredictionResult
+
+_MILLISECOND_TO_MICROSECOND = 1000
+_MICROSECOND_TO_NANOSECOND = 1000
+_SECOND_TO_MICROSECOND = 1000000
+
+
+def _unbatch(maybe_keyed_batches: Any):
+  keys, results = maybe_keyed_batches
+  if keys:
+    return zip(keys, results)
+  else:
+    return results
+
+
+class ModelLoader:
+  """Has the ability to load an ML model."""
+  def load_model(self):
+    """Loads an initializes a model for processing."""
+    raise NotImplementedError(type(self))
+
+
+class InferenceRunner:
+  """Implements running inferences for a framework."""
+  def run_inference(self, batch: Any, model: Any) -> List[PredictionResult]:
+    """Runs inferences on a batch of examples and returns a list of Predictions."""
+    raise NotImplementedError(type(self))
+
+
+class MetricsCollector:
+  """A metrics collector that tracks ML related performance and memory usage."""
+  def __init__(self, namespace: str):
+    # Metrics
+    self._inference_counter = beam.metrics.Metrics.counter(
+        namespace, 'num_inferences')
+    self._inference_request_batch_size = beam.metrics.Metrics.distribution(
+        namespace, 'inference_request_batch_size')
+    self._inference_request_batch_byte_size = (
+        beam.metrics.Metrics.distribution(
+            namespace, 'inference_request_batch_byte_size'))
+    # Batch inference latency in microseconds.
+    self._inference_batch_latency_micro_secs = (
+        beam.metrics.Metrics.distribution(
+            namespace, 'inference_batch_latency_micro_secs'))
+    self._model_byte_size = beam.metrics.Metrics.distribution(
+        namespace, 'model_byte_size')
+    # Model load latency in milliseconds.
+    self._load_model_latency_milli_secs = beam.metrics.Metrics.distribution(
+        namespace, 'load_model_latency_milli_secs')
+
+    # Metrics cache
+    self.load_model_latency_milli_secs_cache = None
+    self.model_byte_size_cache = None
+
+  def update_metrics_with_cache(self):
+    if self.load_model_latency_milli_secs_cache is not None:
+      self._load_model_latency_milli_secs.update(
+          self.load_model_latency_milli_secs_cache)
+      self.load_model_latency_milli_secs_cache = None
+    if self.model_byte_size_cache is not None:
+      self._model_byte_size.update(self.model_byte_size_cache)
+      self.model_byte_size_cache = None
+
+  def update(
+      self,
+      examples_count: int,
+      examples_byte_size: int,
+      latency_micro_secs: int):
+    self._inference_batch_latency_micro_secs.update(latency_micro_secs)
+    self._inference_counter.inc(examples_count)
+    self._inference_request_batch_size.update(examples_count)
+    self._inference_request_batch_byte_size.update(examples_byte_size)
+
+
+class RunInferenceDoFn(beam.DoFn):
+  def __init__(self, model_loader, inference_runner):
+    self._model_loader = model_loader
+    self._inference_runner = inference_runner
+    self._shared_model_handle = shared.Shared()
+    # TODO: Compute a good metrics namespace
+    self._metrics_collector = MetricsCollector('default_namespace')
+    self._clock = _ClockFactory.make_clock()
+    self._model = None
+
+  def _load_model(self):
+    def load():
+      """Function for constructing shared LoadedModel."""
+      memory_before = _get_current_process_memory_in_bytes()
+      start_time = self._clock.get_current_time_in_microseconds()
+      model = self._model_loader.load_model()
+      end_time = self._clock.get_current_time_in_microseconds()
+      memory_after = _get_current_process_memory_in_bytes()
+      self._metrics_collector.load_model_latency_milli_secs_cache = (
+          (end_time - start_time) / _MILLISECOND_TO_MICROSECOND)
+      self._metrics_collector.model_byte_size_cache = (
+          memory_after - memory_before)
+      return model
+
+    return self._shared_model_handle.acquire(load)
+
+  def setup(self):
+    super().setup()
+    self._model = self._load_model()
+
+  def process(self, batch):
+    has_keys = isinstance(batch[0], tuple)
+    start_time = self._clock.get_current_time_in_microseconds()
+    if has_keys:
+      examples = [example for _, example in batch]
+      keys = [key for key, _ in batch]
+    else:
+      examples = batch
+      keys = None
+    inferences = self._inference_runner.run_inference(examples, self._model)
+    inference_latency = self._clock.get_current_time_in_microseconds(
+    ) - start_time
+    num_bytes = sys.getsizeof(batch)
+    num_elements = len(batch)
+    self._metrics_collector.update(
+        num_elements, sys.getsizeof(batch), inference_latency)
+    yield keys, [PredictionResult(e, r) for e, r in zip(examples, inferences)]
+
+
+class RunInferenceImpl(beam.PTransform):
+  def __init__(self, model_loader, inference_runner):
+    self._model_loader = model_loader
+    self._inference_runner = inference_runner
+
+  def expand(self, pcoll: beam.PCollection) -> beam.PCollection:
+    return (
+        pcoll
+        # TODO: Hook into the batching DoFn APIs.
+        | beam.BatchElements()
+        | beam.ParDo(
+            RunInferenceDoFn(self._model_loader, self._inference_runner))
+        | beam.FlatMap(_unbatch))
+
+
+def _is_darwin() -> bool:
+  return sys.platform == 'darwin'
+
+
+def _get_current_process_memory_in_bytes():
+  """Returns memory usage in bytes."""
+
+  if resource is not None:
+    usage = resource.getrusage(resource.RUSAGE_SELF).ru_maxrss
+    if _is_darwin():
+      return usage
+    return usage * 1024
+  else:
+    logging.warning(
+        'Resource module is not available for current platform, '
+        'memory usage cannot be fetched.')
+  return 0
+
+
+def _is_windows() -> bool:
+  return platform.system() == 'Windows' or os.name == 'nt'
+
+
+def _is_cygwin() -> bool:
+  return platform.system().startswith('CYGWIN_NT')
+
+
+class _Clock(object):

Review comment:
       Do we want a test for Clock?

##########
File path: sdks/python/apache_beam/ml/inference/base.py
##########
@@ -0,0 +1,194 @@
+import logging
+import platform
+import resource
+import sys
+import time
+from typing import Any
+from typing import List
+
+import apache_beam as beam
+from apache_beam.utils import shared
+from apache_beam.ml.inference.apis import PredictionResult
+
+_MILLISECOND_TO_MICROSECOND = 1000
+_MICROSECOND_TO_NANOSECOND = 1000
+_SECOND_TO_MICROSECOND = 1000000
+
+
+def _unbatch(maybe_keyed_batches: Any):
+  keys, results = maybe_keyed_batches
+  if keys:
+    return zip(keys, results)
+  else:
+    return results
+
+
+class ModelLoader:
+  """Has the ability to load an ML model."""
+  def load_model(self):

Review comment:
       ```suggestion
     def load_model(self) -> InferenceRunner:
   ```

##########
File path: sdks/python/apache_beam/ml/inference/base.py
##########
@@ -0,0 +1,194 @@
+import logging
+import platform
+import resource
+import sys
+import time
+from typing import Any
+from typing import List
+
+import apache_beam as beam
+from apache_beam.utils import shared
+from apache_beam.ml.inference.apis import PredictionResult
+
+_MILLISECOND_TO_MICROSECOND = 1000
+_MICROSECOND_TO_NANOSECOND = 1000
+_SECOND_TO_MICROSECOND = 1000000
+
+
+def _unbatch(maybe_keyed_batches: Any):
+  keys, results = maybe_keyed_batches
+  if keys:
+    return zip(keys, results)
+  else:
+    return results
+
+
+class ModelLoader:
+  """Has the ability to load an ML model."""
+  def load_model(self):
+    """Loads an initializes a model for processing."""
+    raise NotImplementedError(type(self))
+
+
+class InferenceRunner:
+  """Implements running inferences for a framework."""
+  def run_inference(self, batch: Any, model: Any) -> List[PredictionResult]:
+    """Runs inferences on a batch of examples and returns a list of Predictions."""
+    raise NotImplementedError(type(self))
+
+
+class MetricsCollector:
+  """A metrics collector that tracks ML related performance and memory usage."""
+  def __init__(self, namespace: str):
+    # Metrics
+    self._inference_counter = beam.metrics.Metrics.counter(
+        namespace, 'num_inferences')
+    self._inference_request_batch_size = beam.metrics.Metrics.distribution(
+        namespace, 'inference_request_batch_size')
+    self._inference_request_batch_byte_size = (
+        beam.metrics.Metrics.distribution(
+            namespace, 'inference_request_batch_byte_size'))
+    # Batch inference latency in microseconds.
+    self._inference_batch_latency_micro_secs = (
+        beam.metrics.Metrics.distribution(
+            namespace, 'inference_batch_latency_micro_secs'))
+    self._model_byte_size = beam.metrics.Metrics.distribution(
+        namespace, 'model_byte_size')
+    # Model load latency in milliseconds.
+    self._load_model_latency_milli_secs = beam.metrics.Metrics.distribution(
+        namespace, 'load_model_latency_milli_secs')
+
+    # Metrics cache
+    self.load_model_latency_milli_secs_cache = None
+    self.model_byte_size_cache = None
+
+  def update_metrics_with_cache(self):
+    if self.load_model_latency_milli_secs_cache is not None:
+      self._load_model_latency_milli_secs.update(
+          self.load_model_latency_milli_secs_cache)
+      self.load_model_latency_milli_secs_cache = None
+    if self.model_byte_size_cache is not None:
+      self._model_byte_size.update(self.model_byte_size_cache)
+      self.model_byte_size_cache = None
+
+  def update(
+      self,
+      examples_count: int,
+      examples_byte_size: int,
+      latency_micro_secs: int):
+    self._inference_batch_latency_micro_secs.update(latency_micro_secs)
+    self._inference_counter.inc(examples_count)
+    self._inference_request_batch_size.update(examples_count)
+    self._inference_request_batch_byte_size.update(examples_byte_size)
+
+
+class RunInferenceDoFn(beam.DoFn):
+  def __init__(self, model_loader, inference_runner):
+    self._model_loader = model_loader
+    self._inference_runner = inference_runner
+    self._shared_model_handle = shared.Shared()
+    # TODO: Compute a good metrics namespace
+    self._metrics_collector = MetricsCollector('default_namespace')
+    self._clock = _ClockFactory.make_clock()
+    self._model = None
+
+  def _load_model(self):
+    def load():
+      """Function for constructing shared LoadedModel."""
+      memory_before = _get_current_process_memory_in_bytes()
+      start_time = self._clock.get_current_time_in_microseconds()
+      model = self._model_loader.load_model()
+      end_time = self._clock.get_current_time_in_microseconds()
+      memory_after = _get_current_process_memory_in_bytes()
+      self._metrics_collector.load_model_latency_milli_secs_cache = (
+          (end_time - start_time) / _MILLISECOND_TO_MICROSECOND)
+      self._metrics_collector.model_byte_size_cache = (
+          memory_after - memory_before)
+      return model
+
+    return self._shared_model_handle.acquire(load)
+
+  def setup(self):
+    super().setup()
+    self._model = self._load_model()
+
+  def process(self, batch):
+    has_keys = isinstance(batch[0], tuple)
+    start_time = self._clock.get_current_time_in_microseconds()
+    if has_keys:
+      examples = [example for _, example in batch]
+      keys = [key for key, _ in batch]
+    else:
+      examples = batch
+      keys = None
+    inferences = self._inference_runner.run_inference(examples, self._model)
+    inference_latency = self._clock.get_current_time_in_microseconds(
+    ) - start_time
+    num_bytes = sys.getsizeof(batch)
+    num_elements = len(batch)
+    self._metrics_collector.update(
+        num_elements, sys.getsizeof(batch), inference_latency)
+    yield keys, [PredictionResult(e, r) for e, r in zip(examples, inferences)]
+
+
+class RunInferenceImpl(beam.PTransform):
+  def __init__(self, model_loader, inference_runner):
+    self._model_loader = model_loader
+    self._inference_runner = inference_runner
+
+  def expand(self, pcoll: beam.PCollection) -> beam.PCollection:
+    return (
+        pcoll
+        # TODO: Hook into the batching DoFn APIs.
+        | beam.BatchElements()
+        | beam.ParDo(
+            RunInferenceDoFn(self._model_loader, self._inference_runner))
+        | beam.FlatMap(_unbatch))
+
+
+def _is_darwin() -> bool:
+  return sys.platform == 'darwin'
+
+
+def _get_current_process_memory_in_bytes():
+  """Returns memory usage in bytes."""
+
+  if resource is not None:
+    usage = resource.getrusage(resource.RUSAGE_SELF).ru_maxrss
+    if _is_darwin():
+      return usage
+    return usage * 1024
+  else:
+    logging.warning(
+        'Resource module is not available for current platform, '
+        'memory usage cannot be fetched.')
+  return 0
+
+
+def _is_windows() -> bool:

Review comment:
       Do we want test for OS?

##########
File path: sdks/python/apache_beam/ml/inference/base.py
##########
@@ -0,0 +1,194 @@
+import logging
+import platform
+import resource
+import sys
+import time
+from typing import Any
+from typing import List
+
+import apache_beam as beam
+from apache_beam.utils import shared
+from apache_beam.ml.inference.apis import PredictionResult
+
+_MILLISECOND_TO_MICROSECOND = 1000
+_MICROSECOND_TO_NANOSECOND = 1000
+_SECOND_TO_MICROSECOND = 1000000
+
+
+def _unbatch(maybe_keyed_batches: Any):
+  keys, results = maybe_keyed_batches
+  if keys:
+    return zip(keys, results)
+  else:
+    return results
+
+
+class ModelLoader:
+  """Has the ability to load an ML model."""
+  def load_model(self):
+    """Loads an initializes a model for processing."""
+    raise NotImplementedError(type(self))
+
+
+class InferenceRunner:
+  """Implements running inferences for a framework."""
+  def run_inference(self, batch: Any, model: Any) -> List[PredictionResult]:
+    """Runs inferences on a batch of examples and returns a list of Predictions."""
+    raise NotImplementedError(type(self))
+
+
+class MetricsCollector:
+  """A metrics collector that tracks ML related performance and memory usage."""
+  def __init__(self, namespace: str):
+    # Metrics
+    self._inference_counter = beam.metrics.Metrics.counter(
+        namespace, 'num_inferences')
+    self._inference_request_batch_size = beam.metrics.Metrics.distribution(
+        namespace, 'inference_request_batch_size')
+    self._inference_request_batch_byte_size = (
+        beam.metrics.Metrics.distribution(
+            namespace, 'inference_request_batch_byte_size'))
+    # Batch inference latency in microseconds.
+    self._inference_batch_latency_micro_secs = (
+        beam.metrics.Metrics.distribution(
+            namespace, 'inference_batch_latency_micro_secs'))
+    self._model_byte_size = beam.metrics.Metrics.distribution(
+        namespace, 'model_byte_size')
+    # Model load latency in milliseconds.
+    self._load_model_latency_milli_secs = beam.metrics.Metrics.distribution(
+        namespace, 'load_model_latency_milli_secs')
+
+    # Metrics cache
+    self.load_model_latency_milli_secs_cache = None
+    self.model_byte_size_cache = None
+
+  def update_metrics_with_cache(self):
+    if self.load_model_latency_milli_secs_cache is not None:
+      self._load_model_latency_milli_secs.update(
+          self.load_model_latency_milli_secs_cache)
+      self.load_model_latency_milli_secs_cache = None
+    if self.model_byte_size_cache is not None:
+      self._model_byte_size.update(self.model_byte_size_cache)
+      self.model_byte_size_cache = None
+
+  def update(
+      self,
+      examples_count: int,
+      examples_byte_size: int,
+      latency_micro_secs: int):
+    self._inference_batch_latency_micro_secs.update(latency_micro_secs)
+    self._inference_counter.inc(examples_count)
+    self._inference_request_batch_size.update(examples_count)
+    self._inference_request_batch_byte_size.update(examples_byte_size)
+
+
+class RunInferenceDoFn(beam.DoFn):
+  def __init__(self, model_loader, inference_runner):
+    self._model_loader = model_loader
+    self._inference_runner = inference_runner
+    self._shared_model_handle = shared.Shared()
+    # TODO: Compute a good metrics namespace
+    self._metrics_collector = MetricsCollector('default_namespace')
+    self._clock = _ClockFactory.make_clock()
+    self._model = None
+
+  def _load_model(self):
+    def load():
+      """Function for constructing shared LoadedModel."""
+      memory_before = _get_current_process_memory_in_bytes()
+      start_time = self._clock.get_current_time_in_microseconds()
+      model = self._model_loader.load_model()
+      end_time = self._clock.get_current_time_in_microseconds()
+      memory_after = _get_current_process_memory_in_bytes()
+      self._metrics_collector.load_model_latency_milli_secs_cache = (
+          (end_time - start_time) / _MILLISECOND_TO_MICROSECOND)
+      self._metrics_collector.model_byte_size_cache = (
+          memory_after - memory_before)
+      return model
+
+    return self._shared_model_handle.acquire(load)
+
+  def setup(self):
+    super().setup()
+    self._model = self._load_model()
+
+  def process(self, batch):
+    has_keys = isinstance(batch[0], tuple)
+    start_time = self._clock.get_current_time_in_microseconds()
+    if has_keys:
+      examples = [example for _, example in batch]
+      keys = [key for key, _ in batch]
+    else:
+      examples = batch
+      keys = None
+    inferences = self._inference_runner.run_inference(examples, self._model)
+    inference_latency = self._clock.get_current_time_in_microseconds(
+    ) - start_time
+    num_bytes = sys.getsizeof(batch)
+    num_elements = len(batch)
+    self._metrics_collector.update(
+        num_elements, sys.getsizeof(batch), inference_latency)
+    yield keys, [PredictionResult(e, r) for e, r in zip(examples, inferences)]
+
+
+class RunInferenceImpl(beam.PTransform):
+  def __init__(self, model_loader, inference_runner):
+    self._model_loader = model_loader
+    self._inference_runner = inference_runner
+
+  def expand(self, pcoll: beam.PCollection) -> beam.PCollection:
+    return (
+        pcoll
+        # TODO: Hook into the batching DoFn APIs.

Review comment:
       Let's file a JIRA for this

##########
File path: sdks/python/apache_beam/ml/inference/base.py
##########
@@ -0,0 +1,194 @@
+import logging
+import platform
+import resource
+import sys
+import time
+from typing import Any
+from typing import List
+
+import apache_beam as beam
+from apache_beam.utils import shared
+from apache_beam.ml.inference.apis import PredictionResult
+
+_MILLISECOND_TO_MICROSECOND = 1000
+_MICROSECOND_TO_NANOSECOND = 1000
+_SECOND_TO_MICROSECOND = 1000000
+
+
+def _unbatch(maybe_keyed_batches: Any):
+  keys, results = maybe_keyed_batches
+  if keys:
+    return zip(keys, results)
+  else:
+    return results
+
+
+class ModelLoader:
+  """Has the ability to load an ML model."""
+  def load_model(self):
+    """Loads an initializes a model for processing."""
+    raise NotImplementedError(type(self))
+
+
+class InferenceRunner:
+  """Implements running inferences for a framework."""
+  def run_inference(self, batch: Any, model: Any) -> List[PredictionResult]:
+    """Runs inferences on a batch of examples and returns a list of Predictions."""
+    raise NotImplementedError(type(self))
+
+
+class MetricsCollector:
+  """A metrics collector that tracks ML related performance and memory usage."""
+  def __init__(self, namespace: str):
+    # Metrics
+    self._inference_counter = beam.metrics.Metrics.counter(
+        namespace, 'num_inferences')
+    self._inference_request_batch_size = beam.metrics.Metrics.distribution(
+        namespace, 'inference_request_batch_size')
+    self._inference_request_batch_byte_size = (
+        beam.metrics.Metrics.distribution(
+            namespace, 'inference_request_batch_byte_size'))
+    # Batch inference latency in microseconds.
+    self._inference_batch_latency_micro_secs = (
+        beam.metrics.Metrics.distribution(
+            namespace, 'inference_batch_latency_micro_secs'))
+    self._model_byte_size = beam.metrics.Metrics.distribution(
+        namespace, 'model_byte_size')
+    # Model load latency in milliseconds.
+    self._load_model_latency_milli_secs = beam.metrics.Metrics.distribution(
+        namespace, 'load_model_latency_milli_secs')
+

Review comment:
       What about `num_failed_inferences`, `num_model_loads`, `num_model_versions`, and `curr_model_version`? I think the latter two are relevant to side-inputs, which we don't have yet.




-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org