def create_pipeline(pipeline_name: Text,
pipeline_root: Text,
data_root_uri: data_types.RuntimeParameter,
train_steps: data_types.RuntimeParameter,
eval_steps: data_types.RuntimeParameter,
ai_platform_training_args: Dict[Text, Text],
ai_platform_serving_args: Dict[Text, Text],
beam_pipeline_args: List[Text],
enable_cache: Optional[bool] = False) -> pipeline.Pipeline:
"""Trains and deploys the Covertype classifier."""
# Brings data into the pipeline and splits the data into training and eval splits
examples = external_input(data_root_uri)
output_config = example_gen_pb2.Output(
split_config=example_gen_pb2.SplitConfig(splits=[
example_gen_pb2.SplitConfig.Split(name='train', hash_buckets=4),
example_gen_pb2.SplitConfig.Split(name='eval', hash_buckets=1)
]))
generate_examples = CsvExampleGen(input=examples)
# Computes statistics over data for visualization and example validation.
generate_statistics = StatisticsGen(
examples=generate_examples.outputs.examples)
# Import a user-provided schema
import_schema = ImporterNode(instance_name='import_user_schema',
source_uri=SCHEMA_FOLDER,
artifact_type=Schema)
# Generates schema based on statistics files.Even though, we use user-provided schema
# we still want to generate the schema of the newest data for tracking and comparison
infer_schema = SchemaGen(statistics=generate_statistics.outputs.statistics)
# Performs anomaly detection based on statistics and data schema.
validate_stats = ExampleValidator(
statistics=generate_statistics.outputs.statistics,
schema=import_schema.outputs.result)
# Performs transformations and feature engineering in training and serving.
transform = Transform(examples=generate_examples.outputs.examples,
schema=import_schema.outputs.result,
module_file=TRANSFORM_MODULE_FILE)
# Trains the model using a user provided trainer function.
train = Trainer(
custom_executor_spec=executor_spec.ExecutorClassSpec(
ai_platform_trainer_executor.GenericExecutor),
# custom_executor_spec=executor_spec.ExecutorClassSpec(trainer_executor.GenericExecutor),
module_file=TRAIN_MODULE_FILE,
transformed_examples=transform.outputs.transformed_examples,
schema=import_schema.outputs.result,
transform_graph=transform.outputs.transform_graph,
train_args={'num_steps': train_steps},
eval_args={'num_steps': eval_steps},
custom_config={'ai_platform_training_args': ai_platform_training_args})
# Get the latest blessed model for model validation.
resolve = ResolverNode(instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
# Uses TFMA to compute a evaluation statistics over features of a model.
accuracy_threshold = tfma.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(lower_bound={'value': 0.5},
upper_bound={'value':
0.99}),
change_threshold=tfma.GenericChangeThreshold(
absolute={'value': 0.0001},
direction=tfma.MetricDirection.HIGHER_IS_BETTER),
)
metrics_specs = tfma.MetricsSpec(metrics=[
tfma.MetricConfig(class_name='SparseCategoricalAccuracy',
threshold=accuracy_threshold),
tfma.MetricConfig(class_name='ExampleCount')
])
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(label_key='Cover_Type')],
metrics_specs=[metrics_specs],
slicing_specs=[
tfma.SlicingSpec(),
tfma.SlicingSpec(feature_keys=['Wilderness_Area'])
])
analyze = Evaluator(examples=generate_examples.outputs.examples,
model=train.outputs.model,
baseline_model=resolve.outputs.model,
eval_config=eval_config)
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
deploy = Pusher(
model=train.outputs['model'],
model_blessing=analyze.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=os.path.join(str(pipeline.ROOT_PARAMETER),
'model_serving'))))
#deploy = Pusher(
# custom_executor_spec=executor_spec.ExecutorClassSpec(
# ai_platform_pusher_executor.Executor),
# model=train.outputs.model,
# model_blessing=validate.outputs.blessing,
# custom_config={'ai_platform_serving_args': ai_platform_serving_args})
return pipeline.Pipeline(pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
generate_examples, generate_statistics,
import_schema, infer_schema, validate_stats,
transform, train, resolve, analyze, deploy
],
enable_cache=enable_cache,
beam_pipeline_args=beam_pipeline_args)
def _create_pipeline():
"""Implements the chicago taxi pipeline with TFX."""
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input_base=_data_root)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'])
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=_taxi_module_file)
# Uses user-provided Python function that implements a model.
trainer = Trainer(module_file=_taxi_module_file,
examples=transform.outputs['transformed_examples'],
schema=schema_gen.outputs['schema'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=10000),
eval_args=trainer_pb2.EvalArgs(num_steps=5000))
# Uses TFMA to compute a evaluation statistics over features of a model.
evaluator = Evaluator(
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
feature_slicing_spec=evaluator_pb2.FeatureSlicingSpec(specs=[
evaluator_pb2.SingleSlicingSpec(
column_for_slicing=['trip_start_hour'])
]))
# Performs quality validation of a candidate model (compared to a baseline).
model_validator = ModelValidator(examples=example_gen.outputs['examples'],
model=trainer.outputs['model'])
# This custom component serves as a bridge between pipeline and human model
# reviewers to enable review-and-push workflow in model development cycle. It
# utilizes Slack API to send message to user-defined Slack channel with model
# URI info and wait for go / no-go decision from the same Slack channel:
# * To approve the model, users need to reply the thread sent out by the bot
# started by SlackComponent with 'lgtm' or 'approve'.
# * To reject the model, users need to reply the thread sent out by the bot
# started by SlackComponent with 'decline' or 'reject'.
slack_validator = SlackComponent(
model=trainer.outputs['model'],
model_blessing=model_validator.outputs['blessing'],
slack_token=_slack_token,
slack_channel_id=_slack_channel_id,
timeout_sec=3600,
)
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher = Pusher(model=trainer.outputs['model'],
model_blessing=slack_validator.outputs['slack_blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=_serving_model_dir)))
return pipeline.Pipeline(
pipeline_name=_pipeline_name,
pipeline_root=_pipeline_root,
components=[
example_gen, statistics_gen, schema_gen, example_validator,
transform, trainer, evaluator, model_validator, slack_validator,
pusher
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
_metadata_db_root),
)
def create_test_pipeline():
"""Builds an Iris example pipeline with slight changes."""
pipeline_name = "iris"
iris_root = "iris_root"
serving_model_dir = os.path.join(iris_root, "serving_model", pipeline_name)
tfx_root = "tfx_root"
data_path = os.path.join(tfx_root, "data_path")
pipeline_root = os.path.join(tfx_root, "pipelines", pipeline_name)
example_gen = CsvExampleGen(input_base=data_path)
statistics_gen = StatisticsGen(examples=example_gen.outputs["examples"])
importer = ImporterNode(
instance_name="my_importer",
source_uri="m/y/u/r/i",
properties={
"split_names": "['train', 'eval']",
},
custom_properties={
"int_custom_property": 42,
"str_custom_property": "42",
},
artifact_type=standard_artifacts.Examples)
another_statistics_gen = StatisticsGen(
examples=importer.outputs["result"],
instance_name="another_statistics_gen")
schema_gen = SchemaGen(statistics=statistics_gen.outputs["statistics"])
example_validator = ExampleValidator(
statistics=statistics_gen.outputs["statistics"],
schema=schema_gen.outputs["schema"])
trainer = Trainer(
# Use RuntimeParameter as module_file to test out RuntimeParameter in
# compiler.
module_file=data_types.RuntimeParameter(
name="module_file",
default=os.path.join(iris_root, "iris_utils.py"),
ptype=str),
custom_executor_spec=executor_spec.ExecutorClassSpec(GenericExecutor),
examples=example_gen.outputs["examples"],
schema=schema_gen.outputs["schema"],
train_args=trainer_pb2.TrainArgs(num_steps=2000),
# Attaching `TrainerArgs` as platform config is not sensible practice,
# but is only for testing purpose.
eval_args=trainer_pb2.EvalArgs(num_steps=5)).with_platform_config(
config=trainer_pb2.TrainArgs(num_steps=2000))
model_resolver = ResolverNode(
instance_name="latest_blessed_model_resolver",
resolver_class=latest_blessed_model_resolver.LatestBlessedModelResolver,
model=Channel(
type=standard_artifacts.Model, producer_component_id=trainer.id),
model_blessing=Channel(type=standard_artifacts.ModelBlessing))
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(signature_name="eval")],
slicing_specs=[tfma.SlicingSpec()],
metrics_specs=[
tfma.MetricsSpec(
thresholds={
"sparse_categorical_accuracy":
tfma.config.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={"value": 0.6}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={"value": -1e-10}))
})
])
evaluator = Evaluator(
examples=example_gen.outputs["examples"],
model=trainer.outputs["model"],
baseline_model=model_resolver.outputs["model"],
eval_config=eval_config)
pusher = Pusher(
model=trainer.outputs["model"],
model_blessing=evaluator.outputs["blessing"],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen,
statistics_gen,
another_statistics_gen,
importer,
schema_gen,
example_validator,
trainer,
model_resolver,
evaluator,
pusher,
],
enable_cache=True,
beam_pipeline_args=["--my_testing_beam_pipeline_args=foo"],
# Attaching `TrainerArgs` as platform config is not sensible practice,
# but is only for testing purpose.
platform_config=trainer_pb2.TrainArgs(num_steps=2000),
execution_mode=pipeline.ExecutionMode.SYNC)
def create_e2e_components(
pipeline_root: Text,
csv_input_location: Text,
transform_module: Text,
trainer_module: Text,
) -> List[BaseComponent]:
"""Creates components for a simple Chicago Taxi TFX pipeline for testing.
Args:
pipeline_root: The root of the pipeline output.
csv_input_location: The location of the input data directory.
transform_module: The location of the transform module file.
trainer_module: The location of the trainer module file.
Returns:
A list of TFX components that constitutes an end-to-end test pipeline.
"""
examples = dsl_utils.csv_input(csv_input_location)
example_gen = CsvExampleGen(input=examples)
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=False)
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=transform_module)
latest_model_resolver = ResolverNode(
instance_name='latest_model_resolver',
resolver_class=latest_artifacts_resolver.LatestArtifactsResolver,
latest_model=Channel(type=Model))
trainer = Trainer(
transformed_examples=transform.outputs['transformed_examples'],
schema=schema_gen.outputs['schema'],
base_model=latest_model_resolver.outputs['latest_model'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=10),
eval_args=trainer_pb2.EvalArgs(num_steps=5),
module_file=trainer_module,
)
evaluator = Evaluator(
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
feature_slicing_spec=evaluator_pb2.FeatureSlicingSpec(specs=[
evaluator_pb2.SingleSlicingSpec(
column_for_slicing=['trip_start_hour'])
]))
model_validator = ModelValidator(examples=example_gen.outputs['examples'],
model=trainer.outputs['model'])
pusher = Pusher(
model=trainer.outputs['model'],
model_blessing=model_validator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=os.path.join(pipeline_root, 'model_serving'))))
return [
example_gen, statistics_gen, schema_gen, example_validator, transform,
latest_model_resolver, trainer, evaluator, model_validator, pusher
]
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
module_file: Text, serving_model_dir: Text,
metadata_path: Text,
beam_pipeline_args: List[Text]) -> pipeline.Pipeline:
"""Implements the chicago taxi pipeline with TFX."""
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input_base=data_root)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=True)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=module_file)
# Uses user-provided Python function that implements a model using TF-Learn.
trainer = Trainer(module_file=module_file,
examples=transform.outputs['transformed_examples'],
transform_graph=transform.outputs['transform_graph'],
schema=schema_gen.outputs['schema'],
train_args=trainer_pb2.TrainArgs(num_steps=1000),
eval_args=trainer_pb2.EvalArgs(num_steps=150))
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(
type=ModelBlessing)).with_id('latest_blessed_model_resolver')
# Uses TFMA to compute a evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[
tfma.ModelSpec(signature_name='serving_default',
label_key='tips_xf',
preprocessing_function_names=['tft_layer'])
],
slicing_specs=[tfma.SlicingSpec()],
metrics_specs=[
tfma.MetricsSpec(metrics=[
tfma.MetricConfig(
class_name='BinaryAccuracy',
threshold=tfma.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.6}),
# Change threshold will be ignored if there is no
# baseline model resolved from MLMD (first run).
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10})))
])
])
evaluator = Evaluator(examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
eval_config=eval_config)
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher = Pusher(model=trainer.outputs['model'],
model_blessing=evaluator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen,
statistics_gen,
schema_gen,
example_validator,
transform,
trainer,
model_resolver,
evaluator,
pusher,
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
beam_pipeline_args=beam_pipeline_args)
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
module_file: Text, serving_model_dir: Text,
metadata_path: Text,
beam_pipeline_args: List[Text]) -> pipeline.Pipeline:
"""Implements the chicago taxi pipeline with TFX."""
# Parametrize data root so it can be replaced on runtime. See the
# "Passing Parameters when triggering dags" section of
# https://airflow.apache.org/docs/apache-airflow/stable/dag-run.html
# for more details.
data_root_runtime = data_types.RuntimeParameter('data_root',
ptype=str,
default=data_root)
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input_base=data_root_runtime)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=False)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=module_file)
# Uses user-provided Python function that implements a model using TF-Learn.
trainer = Trainer(
module_file=module_file,
custom_executor_spec=executor_spec.ExecutorClassSpec(Executor),
transformed_examples=transform.outputs['transformed_examples'],
schema=schema_gen.outputs['schema'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=10000),
eval_args=trainer_pb2.EvalArgs(num_steps=5000))
# Get the latest blessed model for model validation.
model_resolver = resolver.Resolver(
strategy_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(
type=ModelBlessing)).with_id('latest_blessed_model_resolver')
# Uses TFMA to compute a evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(signature_name='eval')],
slicing_specs=[
tfma.SlicingSpec(),
tfma.SlicingSpec(feature_keys=['trip_start_hour'])
],
metrics_specs=[
tfma.MetricsSpec(
thresholds={
'accuracy':
tfma.config.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.6}),
# Change threshold will be ignored if there is no
# baseline model resolved from MLMD (first run).
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10}))
})
])
evaluator = Evaluator(examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
eval_config=eval_config)
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher = Pusher(model=trainer.outputs['model'],
model_blessing=evaluator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen, statistics_gen, schema_gen, example_validator,
transform, trainer, model_resolver, evaluator, pusher
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
beam_pipeline_args=beam_pipeline_args)
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
module_file: Text, serving_model_dir: Text,
metadata_path: Text,
beam_pipeline_args: List[Text]) -> pipeline.Pipeline:
"""Implements the Bert classication on Cola dataset pipline with TFX."""
input_config = example_gen_pb2.Input(splits=[
example_gen_pb2.Input.Split(name='train', pattern='train/*'),
example_gen_pb2.Input.Split(name='eval', pattern='validation/*')
])
examples = external_input(data_root)
# Brings data into the pipline
example_gen = CsvExampleGen(input=examples, input_config=input_config)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=True)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=module_file)
# Uses user-provided Python function that trains a model using TF-Learn.
trainer = Trainer(
module_file=module_file,
custom_executor_spec=executor_spec.ExecutorClassSpec(GenericExecutor),
examples=transform.outputs['transformed_examples'],
transform_graph=transform.outputs['transform_graph'],
schema=schema_gen.outputs['schema'],
# Adjust these steps when training on the full dataset.
train_args=trainer_pb2.TrainArgs(num_steps=2),
eval_args=trainer_pb2.EvalArgs(num_steps=1))
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
# Uses TFMA to compute an evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(label_key='label')],
slicing_specs=[tfma.SlicingSpec()],
metrics_specs=[
tfma.MetricsSpec(metrics=[
tfma.MetricConfig(
class_name='SparseCategoricalAccuracy',
threshold=tfma.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
# Adjust the threshold when training on the
# full dataset.
lower_bound={'value': 0.5}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-2})))
])
])
evaluator = Evaluator(
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
# Change threshold will be ignored if there is no baseline (first run).
eval_config=eval_config)
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher = Pusher(model=trainer.outputs['model'],
model_blessing=evaluator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
components = [
example_gen,
statistics_gen,
schema_gen,
example_validator,
transform,
trainer,
model_resolver,
evaluator,
pusher,
]
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=components,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
enable_cache=True,
beam_pipeline_args=beam_pipeline_args,
)
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
module_file: Text, serving_model_dir: Text,
metadata_path: Text,
beam_pipeline_args: List[Text]) -> pipeline.Pipeline:
"""Implements the Iris flowers pipeline with TFX."""
examples = external_input(data_root)
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input=examples)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=True)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# Uses user-provided Python function that implements a model using TF-Learn.
trainer = Trainer(
module_file=module_file,
# GenericExecutor uses `run_fn`, while default estimator based executor
# uses `trainer_fn` instead.
custom_executor_spec=executor_spec.ExecutorClassSpec(GenericExecutor),
examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
train_args=trainer_pb2.TrainArgs(num_steps=2000),
eval_args=trainer_pb2.EvalArgs(num_steps=5))
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
# Uses TFMA to compute an evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(signature_name='eval')],
slicing_specs=[tfma.SlicingSpec()],
metrics_specs=[
tfma.MetricsSpec(
thresholds={
'sparse_categorical_accuracy':
tfma.config.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.6}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10}))
})
])
evaluator = Evaluator(
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
# Change threshold will be ignored if there is no baseline (first run).
eval_config=eval_config)
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher = Pusher(model=trainer.outputs['model'],
model_blessing=evaluator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen,
statistics_gen,
schema_gen,
example_validator,
trainer,
model_resolver,
evaluator,
pusher,
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
beam_pipeline_args=beam_pipeline_args)
def _set_up_test_pipeline(self):
"""Builds an Iris example pipeline with slight changes."""
pipeline_name = "iris"
iris_root = "iris_root"
serving_model_dir = os.path.join(iris_root, "serving_model",
pipeline_name)
tfx_root = "tfx_root"
data_path = os.path.join(tfx_root, "data_path")
pipeline_root = os.path.join(tfx_root, "pipelines", pipeline_name)
self.test_pipeline_info = data_types.PipelineInfo(
pipeline_name, iris_root)
example_gen = CsvExampleGen(input=external_input(data_path))
statistics_gen = StatisticsGen(
examples=example_gen.outputs["examples"])
schema_gen = SchemaGen(statistics=statistics_gen.outputs["statistics"],
infer_feature_shape=True)
example_validator = ExampleValidator(
statistics=statistics_gen.outputs["statistics"],
schema=schema_gen.outputs["schema"])
trainer = Trainer(
# Use RuntimeParameter as module_file to test out RuntimeParameter in
# compiler.
module_file=data_types.RuntimeParameter(name="module_file",
default=os.path.join(
iris_root,
"iris_utils.py"),
ptype=str),
custom_executor_spec=executor_spec.ExecutorClassSpec(
GenericExecutor),
examples=example_gen.outputs["examples"],
schema=schema_gen.outputs["schema"],
train_args=trainer_pb2.TrainArgs(num_steps=2000),
eval_args=trainer_pb2.EvalArgs(num_steps=5))
model_resolver = ResolverNode(
instance_name="latest_blessed_model_resolver",
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(signature_name="eval")],
slicing_specs=[tfma.SlicingSpec()],
metrics_specs=[
tfma.MetricsSpec(
thresholds={
"sparse_categorical_accuracy":
tfma.config.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={"value": 0.6}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.
HIGHER_IS_BETTER,
absolute={"value": -1e-10}))
})
])
evaluator = Evaluator(examples=example_gen.outputs["examples"],
model=trainer.outputs["model"],
baseline_model=model_resolver.outputs["model"],
eval_config=eval_config)
pusher = Pusher(model=trainer.outputs["model"],
model_blessing=evaluator.outputs["blessing"],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
self._pipeline = pipeline.Pipeline(pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen,
statistics_gen,
schema_gen,
example_validator,
trainer,
model_resolver,
evaluator,
pusher,
],
enable_cache=True,
beam_pipeline_args=[])
pipeline_name = 'titanic'
airflow_dir = os.environ['AIRFLOW_HOME']
data_dir = os.path.join(airflow_dir, 'dags', 'data')
transform_module = os.path.join(airflow_dir, 'dags', 'transform.py')
train_module = os.path.join(airflow_dir, 'dags', 'trainer.py')
serving_model_dir = os.path.join(airflow_dir, 'serving_model', pipeline_name)
tfx_dir = os.path.join(airflow_dir, 'tfx')
pipeline_dir = os.path.join(tfx_dir, 'pipelines', pipeline_name)
metadata_path = os.path.join(tfx_dir, 'metadata', pipeline_name, 'metadata.db')
"""Implements the Titanic kaggle with TFX."""
example_gen = CsvExampleGen(input=external_input(data_dir))
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=False)
validate_stats = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=transform_module)
trainer = Trainer(
def create_pipeline(
pipeline_name: Text,
pipeline_root: Text,
data_root: Text,
module_file: Text,
ai_platform_training_args: Dict[Text, Text],
ai_platform_serving_args: Dict[Text, Text],
enable_tuning: bool,
beam_pipeline_args: Optional[List[Text]] = None) -> pipeline.Pipeline:
"""Implements the penguin pipeline with TFX and Kubeflow Pipeline.
Args:
pipeline_name: name of the TFX pipeline being created.
pipeline_root: root directory of the pipeline. Should be a valid GCS path.
data_root: uri of the penguin data.
module_file: uri of the module files used in Trainer and Transform
components.
ai_platform_training_args: Args of CAIP training job. Please refer to
https://cloud.google.com/ml-engine/reference/rest/v1/projects.jobs#Job
for detailed description.
ai_platform_serving_args: Args of CAIP model deployment. Please refer to
https://cloud.google.com/ml-engine/reference/rest/v1/projects.models
for detailed description.
enable_tuning: If True, the hyperparameter tuning through CloudTuner is
enabled.
beam_pipeline_args: Optional list of beam pipeline options. Please refer to
https://cloud.google.com/dataflow/docs/guides/specifying-exec-params#setting-other-cloud-dataflow-pipeline-options.
When this argument is not provided, the default is to use GCP
DataflowRunner with 50GB disk size as specified in this function. If an
empty list is passed in, default specified by Beam will be used, which can
be found at
https://cloud.google.com/dataflow/docs/guides/specifying-exec-params#setting-other-cloud-dataflow-pipeline-options
Returns:
A TFX pipeline object.
"""
# Beam args to run data processing on DataflowRunner.
#
# TODO(b/151114974): Remove `disk_size_gb` flag after default is increased.
# TODO(b/151116587): Remove `shuffle_mode` flag after default is changed.
# TODO(b/156874687): Remove `machine_type` after IP addresses are no longer a
# scaling bottleneck.
if beam_pipeline_args is None:
beam_pipeline_args = [
'--runner=DataflowRunner',
'--project=' + _project_id,
'--temp_location=' + os.path.join(_pipeline_root, 'tmp'),
'--region=' + _gcp_region,
# Temporary overrides of defaults.
'--disk_size_gb=50',
'--experiments=shuffle_mode=auto',
'--machine_type=e2-standard-8',
]
# Number of epochs in training.
train_steps = data_types.RuntimeParameter(
name='train_steps',
default=100,
ptype=int,
)
# Number of epochs in evaluation.
eval_steps = data_types.RuntimeParameter(
name='eval_steps',
default=50,
ptype=int,
)
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input_base=data_root)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=True)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=module_file)
# Update ai_platform_training_args if distributed training was enabled.
# Number of worker machines used in distributed training.
worker_count = data_types.RuntimeParameter(
name='worker_count',
default=2,
ptype=int,
)
# Type of worker machines used in distributed training.
worker_type = data_types.RuntimeParameter(
name='worker_type',
default='standard',
ptype=str,
)
local_training_args = copy.deepcopy(ai_platform_training_args)
if FLAGS.distributed_training:
local_training_args.update({
# You can specify the machine types, the number of replicas for workers
# and parameter servers.
# https://cloud.google.com/ml-engine/reference/rest/v1/projects.jobs#ScaleTier
'scaleTier': 'CUSTOM',
'masterType': 'large_model',
'workerType': worker_type,
'parameterServerType': 'standard',
'workerCount': worker_count,
'parameterServerCount': 1,
})
# Tunes the hyperparameters for model training based on user-provided Python
# function. Note that once the hyperparameters are tuned, you can drop the
# Tuner component from pipeline and feed Trainer with tuned hyperparameters.
if enable_tuning:
# The Tuner component launches 1 CAIP Training job for flock management.
# For example, 3 workers (defined by num_parallel_trials) in the flock
# management CAIP Training job, each runs Tuner.Executor.
# Then, 3 CAIP Training Jobs (defined by local_training_args) are invoked
# from each worker in the flock management Job for Trial execution.
tuner = Tuner(
module_file=module_file,
examples=transform.outputs['transformed_examples'],
transform_graph=transform.outputs['transform_graph'],
train_args={'num_steps': train_steps},
eval_args={'num_steps': eval_steps},
tune_args=tuner_pb2.TuneArgs(
# num_parallel_trials=3 means that 3 search loops are
# running in parallel.
# Each tuner may include a distributed training job which can be
# specified in local_training_args above (e.g. 1 PS + 2 workers).
num_parallel_trials=3),
custom_config={
# Configures Cloud AI Platform-specific configs . For details, see
# https://cloud.google.com/ai-platform/training/docs/reference/rest/v1/projects.jobs#traininginput.
ai_platform_trainer_executor.TRAINING_ARGS_KEY:
local_training_args
})
# Uses user-provided Python function that trains a model.
trainer = Trainer(
custom_executor_spec=executor_spec.ExecutorClassSpec(
ai_platform_trainer_executor.GenericExecutor),
module_file=module_file,
examples=transform.outputs['transformed_examples'],
transform_graph=transform.outputs['transform_graph'],
schema=schema_gen.outputs['schema'],
# If Tuner is in the pipeline, Trainer can take Tuner's output
# best_hyperparameters artifact as input and utilize it in the user module
# code.
#
# If there isn't Tuner in the pipeline, either use ImporterNode to import
# a previous Tuner's output to feed to Trainer, or directly use the tuned
# hyperparameters in user module code and set hyperparameters to None
# here.
#
# Example of ImporterNode,
# hparams_importer = ImporterNode(
# instance_name='import_hparams',
# source_uri='path/to/best_hyperparameters.txt',
# artifact_type=HyperParameters)
# ...
# hyperparameters = hparams_importer.outputs['result'],
hyperparameters=(tuner.outputs['best_hyperparameters']
if enable_tuning else None),
train_args={'num_steps': train_steps},
eval_args={'num_steps': eval_steps},
custom_config={
ai_platform_trainer_executor.TRAINING_ARGS_KEY: local_training_args
})
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
# Uses TFMA to compute evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(label_key='species')],
slicing_specs=[tfma.SlicingSpec()],
metrics_specs=[
tfma.MetricsSpec(metrics=[
tfma.MetricConfig(
class_name='SparseCategoricalAccuracy',
threshold=tfma.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.6}),
# Change threshold will be ignored if there is no
# baseline model resolved from MLMD (first run).
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10})))
])
])
evaluator = Evaluator(examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
eval_config=eval_config)
pusher = Pusher(
custom_executor_spec=executor_spec.ExecutorClassSpec(
ai_platform_pusher_executor.Executor),
model=trainer.outputs['model'],
model_blessing=evaluator.outputs['blessing'],
custom_config={
ai_platform_pusher_executor.SERVING_ARGS_KEY:
ai_platform_serving_args
},
)
components = [
example_gen,
statistics_gen,
schema_gen,
example_validator,
transform,
trainer,
model_resolver,
evaluator,
pusher,
]
if enable_tuning:
components.append(tuner)
return pipeline.Pipeline(pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=components,
enable_cache=True,
beam_pipeline_args=beam_pipeline_args)
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
module_file: Text, serving_model_dir: Text,
metadata_path: Text, enable_tuning: bool,
beam_pipeline_args: List[Text]) -> pipeline.Pipeline:
"""Implements the penguin pipeline with TFX."""
examples = external_input(data_root)
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input=examples)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=True)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=module_file)
# Tunes the hyperparameters for model training based on user-provided Python
# function. Note that once the hyperparameters are tuned, you can drop the
# Tuner component from pipeline and feed Trainer with tuned hyperparameters.
if enable_tuning:
tuner = Tuner(module_file=module_file,
examples=transform.outputs['transformed_examples'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=20),
eval_args=trainer_pb2.EvalArgs(num_steps=5))
# Uses user-provided Python function that trains a model.
trainer = Trainer(
module_file=module_file,
custom_executor_spec=executor_spec.ExecutorClassSpec(GenericExecutor),
examples=transform.outputs['transformed_examples'],
transform_graph=transform.outputs['transform_graph'],
schema=schema_gen.outputs['schema'],
# If Tuner is in the pipeline, Trainer can take Tuner's output
# best_hyperparameters artifact as input and utilize it in the user module
# code.
#
# If there isn't Tuner in the pipeline, either use ImporterNode to import
# a previous Tuner's output to feed to Trainer, or directly use the tuned
# hyperparameters in user module code and set hyperparameters to None
# here.
#
# Example of ImporterNode,
# hparams_importer = ImporterNode(
# instance_name='import_hparams',
# source_uri='path/to/best_hyperparameters.txt',
# artifact_type=HyperParameters)
# ...
# hyperparameters = hparams_importer.outputs['result'],
hyperparameters=(tuner.outputs['best_hyperparameters']
if enable_tuning else None),
train_args=trainer_pb2.TrainArgs(num_steps=100),
eval_args=trainer_pb2.EvalArgs(num_steps=5))
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
# Uses TFMA to compute an evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(label_key='species')],
slicing_specs=[tfma.SlicingSpec()],
metrics_specs=[
tfma.MetricsSpec(metrics=[
tfma.MetricConfig(
class_name='SparseCategoricalAccuracy',
threshold=tfma.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.6}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10})))
])
])
evaluator = Evaluator(
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
# Change threshold will be ignored if there is no baseline (first run).
eval_config=eval_config)
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher = Pusher(model=trainer.outputs['model'],
model_blessing=evaluator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
components = [
example_gen,
statistics_gen,
schema_gen,
example_validator,
transform,
trainer,
model_resolver,
evaluator,
pusher,
]
if enable_tuning:
components.append(tuner)
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=components,
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
beam_pipeline_args=beam_pipeline_args)
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
module_file: Text, serving_model_dir: Text,
metadata_path: Text,
worker_parallelism: int) -> pipeline.Pipeline:
"""Implements the chicago taxi pipeline with TFX."""
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input_base=data_root)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=False)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=module_file)
# Uses user-provided Python function that implements a model using TF-Learn.
trainer = Trainer(
module_file=module_file,
transformed_examples=transform.outputs['transformed_examples'],
schema=schema_gen.outputs['schema'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=10000),
eval_args=trainer_pb2.EvalArgs(num_steps=5000))
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
# Uses TFMA to compute a evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(signature_name='eval')],
slicing_specs=[
tfma.SlicingSpec(),
tfma.SlicingSpec(feature_keys=['trip_start_hour'])
],
metrics_specs=[
tfma.MetricsSpec(
thresholds={
'accuracy':
tfma.config.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.6}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10}))
})
])
evaluator = Evaluator(
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
# Change threshold will be ignored if there is no baseline (first run).
eval_config=eval_config)
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher = Pusher(model=trainer.outputs['model'],
model_blessing=evaluator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen, statistics_gen, schema_gen, example_validator,
transform, trainer, model_resolver, evaluator, pusher
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
# LINT.IfChange
beam_pipeline_args=[
# ---------------------------- Beam Args --------------------------- #
'--runner=PortableRunner',
# Points to the job server started in
# setup_beam_on_{flink, spark}.sh
'--job_endpoint=localhost:8099',
'--environment_type=LOOPBACK',
'--sdk_worker_parallelism=%d' % worker_parallelism,
'--experiments=use_loopback_process_worker=True',
# Setting environment_cache_millis to practically infinity enables
# continual reuse of Beam SDK workers, improving performance.
'--environment_cache_millis=1000000',
# TODO(BEAM-7199): Obviate setting this.
'--experiments=pre_optimize=all',
# ------------------------ End of Beam Args ------------------------ #
# ------------------------------------------------------------------ #
# ------------ Flink runner Args (ignored by Spark runner) --------- #
'--parallelism=%d' % worker_parallelism,
# TODO(b/175810858): Obviate setting this.
'--execution_mode_for_batch=BATCH_FORCED',
# -------------------- End of Flink runner Args -------------------- #
],
# LINT.ThenChange(setup/setup_beam_on_spark.sh,
# setup/setup_beam_on_flink.sh)
)
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
module_file: Text, serving_model_dir: Text,
metadata_path: Text,
beam_pipeline_args: List[Text]) -> pipeline.Pipeline:
"""Implements the Iris flowers pipeline with TFX."""
examples = external_input(data_root)
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input=examples)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=True)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=module_file)
# Uses user-provided Python function that trains a model using TF-Learn.
trainer = Trainer(
module_file=module_file,
custom_executor_spec=executor_spec.ExecutorClassSpec(GenericExecutor),
examples=transform.outputs['transformed_examples'],
transform_graph=transform.outputs['transform_graph'],
schema=schema_gen.outputs['schema'],
train_args=trainer_pb2.TrainArgs(num_steps=2000),
eval_args=trainer_pb2.EvalArgs(num_steps=5))
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
# Uses TFMA to compute an evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(label_key='variety')],
slicing_specs=[tfma.SlicingSpec()],
metrics_specs=[
tfma.MetricsSpec(metrics=[
tfma.MetricConfig(
class_name='SparseCategoricalAccuracy',
threshold=tfma.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.6}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10})))
])
])
evaluator = Evaluator(
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
# Change threshold will be ignored if there is no baseline (first run).
eval_config=eval_config)
# Performs infra validation of a candidate model to prevent unservable model
# from being pushed. This config will launch a model server of the latest
# TensorFlow Serving image in a local docker engine.
infra_validator = InfraValidator(
model=trainer.outputs['model'],
examples=example_gen.outputs['examples'],
serving_spec=infra_validator_pb2.ServingSpec(
tensorflow_serving=infra_validator_pb2.TensorFlowServing(
tags=['latest']),
local_docker=infra_validator_pb2.LocalDockerConfig()),
request_spec=infra_validator_pb2.RequestSpec(
tensorflow_serving=infra_validator_pb2.
TensorFlowServingRequestSpec()))
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher = Pusher(model=trainer.outputs['model'],
model_blessing=evaluator.outputs['blessing'],
infra_blessing=infra_validator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen,
statistics_gen,
schema_gen,
example_validator,
transform,
trainer,
model_resolver,
evaluator,
infra_validator,
pusher,
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
beam_pipeline_args=beam_pipeline_args)
def _create_pipeline(pipeline_name: Text, pipeline_root: Text,
training_data_root: Text, inference_data_root: Text,
module_file: Text, metadata_path: Text,
direct_num_workers: int) -> pipeline.Pipeline:
"""Implements the chicago taxi pipeline with TFX."""
training_examples = external_input(training_data_root)
# Brings training data into the pipeline or otherwise joins/converts
# training data.
training_example_gen = CsvExampleGen(input_base=training_examples,
instance_name='training_example_gen')
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(
input_data=training_example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=False)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(examples=training_example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=module_file)
# Uses user-provided Python function that implements a model using TF-Learn.
trainer = Trainer(
module_file=module_file,
transformed_examples=transform.outputs['transformed_examples'],
schema=schema_gen.outputs['schema'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=10000),
eval_args=trainer_pb2.EvalArgs(num_steps=5000))
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
# Uses TFMA to compute a evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(signature_name='eval')],
slicing_specs=[
tfma.SlicingSpec(),
tfma.SlicingSpec(feature_keys=['trip_start_hour'])
],
metrics_specs=[
tfma.MetricsSpec(
thresholds={
'accuracy':
tfma.config.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.6}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10}))
})
])
evaluator = Evaluator(
examples=training_example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
# Change threshold will be ignored if there is no baseline (first run).
eval_config=eval_config)
inference_examples = external_input(inference_data_root)
# Brings inference data into the pipeline.
inference_example_gen = CsvExampleGen(
input_base=inference_examples,
output_config=example_gen_pb2.Output(
split_config=example_gen_pb2.SplitConfig(splits=[
example_gen_pb2.SplitConfig.Split(name='unlabelled',
hash_buckets=100)
])),
instance_name='inference_example_gen')
# Performs offline batch inference over inference examples.
bulk_inferrer = BulkInferrer(
examples=inference_example_gen.outputs['examples'],
model=trainer.outputs['model'],
model_blessing=evaluator.outputs['blessing'],
# Empty data_spec.example_splits will result in using all splits.
data_spec=bulk_inferrer_pb2.DataSpec(),
model_spec=bulk_inferrer_pb2.ModelSpec())
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
training_example_gen, inference_example_gen, statistics_gen,
schema_gen, example_validator, transform, trainer, model_resolver,
evaluator, bulk_inferrer
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
# TODO(b/142684737): The multi-processing API might change.
beam_pipeline_args=['--direct_num_workers=%d' % direct_num_workers])
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
module_file: Text, serving_model_dir: Text,
metadata_path: Text,
direct_num_workers: int) -> pipeline.Pipeline:
"""Implements the chicago taxi pipeline with TFX."""
examples = external_input(data_root)
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input=examples)
# Computes statistics over data for visualization and example validation.
# statistics_gen = StatisticsGen(examples=example_gen.outputs['examples']) # Step 3
# Generates schema based on statistics files.
# infer_schema = SchemaGen( # Step 3
# statistics=statistics_gen.outputs['statistics'], # Step 3
# infer_feature_shape=False) # Step 3
# Performs anomaly detection based on statistics and data schema.
# validate_stats = ExampleValidator( # Step 3
# statistics=statistics_gen.outputs['statistics'], # Step 3
# schema=infer_schema.outputs['schema']) # Step 3
# Performs transformations and feature engineering in training and serving.
# transform = Transform( # Step 4
# examples=example_gen.outputs['examples'], # Step 4
# schema=infer_schema.outputs['schema'], # Step 4
# module_file=module_file) # Step 4
# Uses user-provided Python function that implements a model using TF-Learn.
# trainer = Trainer( # Step 5
# module_file=module_file, # Step 5
# custom_executor_spec=executor_spec.ExecutorClassSpec(GenericExecutor), # Step 5
# examples=transform.outputs['transformed_examples'], # Step 5
# transform_graph=transform.outputs['transform_graph'], # Step 5
# schema=infer_schema.outputs['schema'], # Step 5
# train_args=trainer_pb2.TrainArgs(num_steps=10000), # Step 5
# eval_args=trainer_pb2.EvalArgs(num_steps=5000)) # Step 5
# Get the latest blessed model for model validation.
# model_resolver = ResolverNode( # Step 6
# instance_name='latest_blessed_model_resolver', # Step 6
# resolver_class=latest_blessed_model_resolver.LatestBlessedModelResolver, # Step 6
# model=Channel(type=Model), # Step 6
# model_blessing=Channel(type=ModelBlessing)) # Step 6
# Uses TFMA to compute a evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
# eval_config = tfma.EvalConfig( # Step 6
# model_specs=[tfma.ModelSpec(label_key='tips')], # Step 6
# slicing_specs=[tfma.SlicingSpec()], # Step 6
# metrics_specs=[ # Step 6
# tfma.MetricsSpec(metrics=[ # Step 6
# tfma.MetricConfig( # Step 6
# class_name='BinaryAccuracy', # Step 6
# threshold=tfma.MetricThreshold( # Step 6
# value_threshold=tfma.GenericValueThreshold( # Step 6
# lower_bound={'value': 0.6}), # Step 6
# change_threshold=tfma.GenericChangeThreshold( # Step 6
# direction=tfma.MetricDirection.HIGHER_IS_BETTER, # Step 6
# absolute={'value': -1e-10}))) # Step 6
# ]) # Step 6
# ]) # Step 6
# model_analyzer = Evaluator( # Step 6
# examples=example_gen.outputs['examples'], # Step 6
# model=trainer.outputs['model'], # Step 6
# baseline_model=model_resolver.outputs['model'], # Step 6
# # Change threshold will be ignored if there is no baseline (first run). # Step 6
# eval_config=eval_config) # Step 6
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
# pusher = Pusher( # Step 7
# model=trainer.outputs['model'], # Step 7
# model_blessing=model_analyzer.outputs['blessing'], # Step 7
# push_destination=pusher_pb2.PushDestination( # Step 7
# filesystem=pusher_pb2.PushDestination.Filesystem( # Step 7
# base_directory=serving_model_dir))) # Step 7
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen,
# statistics_gen, # Step 3
# infer_schema, # Step 3
# validate_stats, # Step 3
# transform, # Step 4
# trainer, # Step 5
# model_resolver, # Step 6
# model_analyzer, # Step 6
# pusher, # Step 7
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
# TODO(b/142684737): The multi-processing API might change.
beam_pipeline_args=['--direct_num_workers=%d' % direct_num_workers])
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
trainer_module_file: Text, evaluator_module_file: Text,
serving_model_dir: Text, metadata_path: Text,
beam_pipeline_args: List[Text]) -> pipeline.Pipeline:
"""Implements the Penguin pipeline with TFX."""
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input_base=data_root)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=True)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# TODO(humichael): Handle applying transformation component in Milestone 3.
# Uses user-provided Python function that trains a model using TF-Learn.
# Num_steps is not provided during evaluation because the scikit-learn model
# loads and evaluates the entire test set at once.
trainer = Trainer(module_file=trainer_module_file,
examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
train_args=trainer_pb2.TrainArgs(num_steps=2000),
eval_args=trainer_pb2.EvalArgs())
# Get the latest blessed model for model validation.
model_resolver = resolver.Resolver(
strategy_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(
type=ModelBlessing)).with_id('latest_blessed_model_resolver')
# Uses TFMA to compute evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(label_key='species')],
slicing_specs=[tfma.SlicingSpec()],
metrics_specs=[
tfma.MetricsSpec(metrics=[
tfma.MetricConfig(
class_name='Accuracy',
threshold=tfma.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.6}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10})))
])
])
evaluator = Evaluator(module_file=evaluator_module_file,
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
eval_config=eval_config)
pusher = Pusher(model=trainer.outputs['model'],
model_blessing=evaluator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen,
statistics_gen,
schema_gen,
example_validator,
trainer,
model_resolver,
evaluator,
pusher,
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
beam_pipeline_args=beam_pipeline_args,
)
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
user_schema_path: Text, module_file: Text,
serving_model_dir: Text, metadata_path: Text,
direct_num_workers: int) -> pipeline.Pipeline:
"""Implements the chicago taxi pipeline with TFX."""
examples = external_input(data_root)
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input=examples)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Import user-provided schema.
user_schema_importer = ImporterNode(
instance_name='import_user_schema',
source_uri=user_schema_path,
artifact_type=Schema)
# Generates schema based on statistics files. Even we use user-provided schema
# in downstream components, we still want to generate the schema of the newest
# data so that user can compare and optionally update the schema to use.
infer_schema = SchemaGen(
statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=False)
# Performs anomaly detection based on statistics and data schema.
validate_stats = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=user_schema_importer.outputs['result'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(
examples=example_gen.outputs['examples'],
schema=user_schema_importer.outputs['result'],
module_file=module_file)
# Uses user-provided Python function that implements a model using TF-Learn.
trainer = Trainer(
module_file=module_file,
transformed_examples=transform.outputs['transformed_examples'],
schema=user_schema_importer.outputs['result'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=10000),
eval_args=trainer_pb2.EvalArgs(num_steps=5000))
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
# Uses TFMA to compute a evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(signature_name='eval')],
slicing_specs=[
tfma.SlicingSpec(),
tfma.SlicingSpec(feature_keys=['trip_start_hour'])
],
metrics_specs=[
tfma.MetricsSpec(
thresholds={
'binary_accuracy':
tfma.config.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.6}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10}))
})
])
model_analyzer = Evaluator(
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
# Change threshold will be ignored if there is no baseline (first run).
eval_config=eval_config)
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher = Pusher(
model=trainer.outputs['model'],
model_blessing=model_analyzer.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen, statistics_gen, user_schema_importer, infer_schema,
validate_stats, transform, trainer, model_resolver, model_analyzer,
pusher
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
# TODO(b/142684737): The multi-processing API might change.
beam_pipeline_args=['--direct_num_workers=%d' % direct_num_workers])
def _create_pipeline(
pipeline_name: Text, pipeline_root: Text
) -> pipeline.Pipeline:
"""Implements the Iris flowers pipeline with TFX."""
examples = external_input(_data_root_param)
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input=examples)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
infer_schema = SchemaGen(
statistics=statistics_gen.outputs['statistics'], infer_feature_shape=True
)
# Performs anomaly detection based on statistics and data schema.
validate_stats = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=infer_schema.outputs['schema']
)
# Performs transformations and feature engineering in training and serving.
transform = Transform(
examples=example_gen.outputs['examples'],
schema=infer_schema.outputs['schema'],
module_file=_module_file_param
)
# Uses user-provided Python function that implements a model using Keras.
trainer = Trainer(
module_file=_module_file_param,
custom_executor_spec=executor_spec.ExecutorClassSpec(GenericExecutor),
examples=transform.outputs['transformed_examples'],
transform_graph=transform.outputs['transform_graph'],
schema=infer_schema.outputs['schema'],
train_args=trainer_pb2.TrainArgs(num_steps=100),
eval_args=trainer_pb2.EvalArgs(num_steps=50)
)
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing)
)
# Uses TFMA to compute an evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
# Note: to compile this successfully you'll need TFMA at >= 0.21.5
eval_config = tfma.EvalConfig(
model_specs=[
tfma.ModelSpec(name='candidate', label_key='variety'),
tfma.ModelSpec(
name='baseline', label_key='variety', is_baseline=True
)
],
slicing_specs=[
tfma.SlicingSpec(),
# Data can be sliced along a feature column. Required by TFMA visualization.
tfma.SlicingSpec(feature_keys=['sepal_length'])],
metrics_specs=[
tfma.MetricsSpec(
metrics=[
tfma.MetricConfig(
class_name='SparseCategoricalAccuracy',
threshold=tfma.config.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.9}
),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10}
)
)
)
]
)
]
)
# Uses TFMA to compute a evaluation statistics over features of a model.
model_analyzer = Evaluator(
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
# Change threshold will be ignored if there is no baseline (first run).
eval_config=eval_config
)
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher = Pusher(
model=trainer.outputs['model'],
model_blessing=model_analyzer.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=os.path.
join(str(pipeline.ROOT_PARAMETER), 'model_serving')
)
)
)
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen, statistics_gen, infer_schema, validate_stats, transform,
trainer, model_resolver, model_analyzer, pusher
],
enable_cache=True,
)
def create_e2e_components(
pipeline_root: Text,
csv_input_location: Text,
transform_module: Text,
trainer_module: Text,
) -> List[BaseComponent]:
"""Creates components for a simple Chicago Taxi TFX pipeline for testing.
Args:
pipeline_root: The root of the pipeline output.
csv_input_location: The location of the input data directory.
transform_module: The location of the transform module file.
trainer_module: The location of the trainer module file.
Returns:
A list of TFX components that constitutes an end-to-end test pipeline.
"""
example_gen = CsvExampleGen(input_base=csv_input_location)
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'])
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=transform_module)
latest_model_resolver = resolver.Resolver(
strategy_class=latest_artifact_strategy.LatestArtifactStrategy,
latest_model=Channel(type=Model)).with_id('latest_model_resolver')
trainer = Trainer(
transformed_examples=transform.outputs['transformed_examples'],
schema=schema_gen.outputs['schema'],
base_model=latest_model_resolver.outputs['latest_model'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=10),
eval_args=trainer_pb2.EvalArgs(num_steps=5),
module_file=trainer_module,
)
# Set the TFMA config for Model Evaluation and Validation.
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(signature_name='eval')],
metrics_specs=[
tfma.MetricsSpec(
metrics=[tfma.MetricConfig(class_name='ExampleCount')],
thresholds={
'accuracy':
tfma.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.5}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10}))
})
],
slicing_specs=[
tfma.SlicingSpec(),
tfma.SlicingSpec(feature_keys=['trip_start_hour'])
])
evaluator = Evaluator(examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
eval_config=eval_config)
infra_validator = InfraValidator(
model=trainer.outputs['model'],
examples=example_gen.outputs['examples'],
serving_spec=infra_validator_pb2.ServingSpec(
tensorflow_serving=infra_validator_pb2.TensorFlowServing(
tags=['latest']),
kubernetes=infra_validator_pb2.KubernetesConfig()),
request_spec=infra_validator_pb2.RequestSpec(
tensorflow_serving=infra_validator_pb2.
TensorFlowServingRequestSpec()))
pusher = Pusher(
model=trainer.outputs['model'],
model_blessing=evaluator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=os.path.join(pipeline_root, 'model_serving'))))
return [
example_gen,
statistics_gen,
schema_gen,
example_validator,
transform,
latest_model_resolver,
trainer,
evaluator,
infra_validator,
pusher,
]
def _create_pipeline(pipeline_name: Text, pipeline_root: Text,
training_data_root: Text, inference_data_root: Text,
module_file: Text,
metadata_path: Text,
direct_num_workers: int) -> pipeline.Pipeline:
"""Implements the chicago taxi pipeline with TFX."""
training_examples = external_input(training_data_root)
# Brings training data into the pipeline or otherwise joins/converts
# training data.
training_example_gen = CsvExampleGen(
input_base=training_examples, instance_name='training_example_gen')
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(
input_data=training_example_gen.outputs['examples'])
# Generates schema based on statistics files.
infer_schema = SchemaGen(
statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=False)
# Performs anomaly detection based on statistics and data schema.
validate_stats = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=infer_schema.outputs['schema'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(
examples=training_example_gen.outputs['examples'],
schema=infer_schema.outputs['schema'],
module_file=module_file)
# Uses user-provided Python function that implements a model using TF-Learn.
trainer = Trainer(
module_file=module_file,
transformed_examples=transform.outputs['transformed_examples'],
schema=infer_schema.outputs['schema'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=10000),
eval_args=trainer_pb2.EvalArgs(num_steps=5000))
# Uses TFMA to compute a evaluation statistics over features of a model.
model_analyzer = Evaluator(
examples=training_example_gen.outputs['examples'],
model_exports=trainer.outputs['model'],
feature_slicing_spec=evaluator_pb2.FeatureSlicingSpec(specs=[
evaluator_pb2.SingleSlicingSpec(
column_for_slicing=['trip_start_hour'])
]))
# Performs quality validation of a candidate model (compared to a baseline).
model_validator = ModelValidator(
examples=training_example_gen.outputs['examples'],
model=trainer.outputs['model'])
inference_examples = external_input(inference_data_root)
# Brings inference data into the pipeline.
inference_example_gen = CsvExampleGen(
input_base=inference_examples,
output_config=example_gen_pb2.Output(
split_config=example_gen_pb2.SplitConfig(
splits=[example_gen_pb2.SplitConfig.Split(
name='unlabelled', hash_buckets=100)])),
instance_name='inference_example_gen')
# Performs offline batch inference over inference examples.
bulk_inferrer = BulkInferrer(
examples=inference_example_gen.outputs['examples'],
model=trainer.outputs['model'],
model_blessing=model_validator.outputs['blessing'],
# Empty data_spec.example_splits will result in using all splits.
data_spec=bulk_inferrer_pb2.DataSpec(),
model_spec=bulk_inferrer_pb2.ModelSpec())
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
training_example_gen, inference_example_gen, statistics_gen,
infer_schema, validate_stats, transform, trainer, model_analyzer,
model_validator, bulk_inferrer
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
# TODO(b/142684737): The multi-processing API might change.
beam_pipeline_args=['--direct_num_workers=%d' % direct_num_workers])
def init_components(data_dir, module_file,
training_steps=TRAIN_STEPS, eval_steps=EVAL_STEPS,
serving_model_dir=None,
ai_platform_training_args=None,
ai_platform_serving_args=None):
if serving_model_dir and ai_platform_serving_args:
raise NotImplementedError(
"Can't set ai_platform_serving_args and serving_model_dir at "
"the same time. Choose one deployment option.")
output = example_gen_pb2.Output(
split_config=example_gen_pb2.SplitConfig(splits=[
example_gen_pb2.SplitConfig.Split(name='train', hash_buckets=99),
example_gen_pb2.SplitConfig.Split(name='eval', hash_buckets=1)
]))
examples = external_input(data_dir)
example_gen = CsvExampleGen(input=examples, output_config=output)
statistics_gen = StatisticsGen(
examples=example_gen.outputs['examples'])
schema_gen = SchemaGen(
statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=False)
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
transform = Transform(
examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=module_file)
training_kwargs = {
"module_file": module_file,
"examples": transform.outputs['transformed_examples'],
"schema": schema_gen.outputs['schema'],
"transform_graph": transform.outputs['transform_graph'],
"train_args": trainer_pb2.TrainArgs(num_steps=training_steps),
"eval_args": trainer_pb2.EvalArgs(num_steps=eval_steps)
}
if ai_platform_training_args:
from tfx.extensions.google_cloud_ai_platform.trainer import executor as ai_platform_trainer_executor
training_kwargs.update({
"custom_executor_spec": executor_spec.ExecutorClassSpec(ai_platform_trainer_executor.GenericExecutor),
"custom_config": {
ai_platform_trainer_executor.TRAINING_ARGS_KEY: ai_platform_training_args}})
else:
training_kwargs.update({"custom_executor_spec": executor_spec.ExecutorClassSpec(GenericExecutor)})
trainer = Trainer(**training_kwargs)
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
eval_config=tfma.EvalConfig(
model_specs=[tfma.ModelSpec(label_key='consumer_disputed')],
slicing_specs=[tfma.SlicingSpec(), tfma.SlicingSpec(feature_keys=['product'])],
metrics_specs=[
tfma.MetricsSpec(metrics=[
tfma.MetricConfig(class_name='BinaryAccuracy'),
tfma.MetricConfig(class_name='ExampleCount'),
tfma.MetricConfig(class_name='AUC')
],
thresholds={
'AUC':
tfma.config.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.65}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': 0.01}))}
)]
)
evaluator = Evaluator(
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
eval_config=eval_config)
pusher_kwargs = {
"model": trainer.outputs['model'],
"model_blessing": evaluator.outputs['blessing'],
}
if ai_platform_serving_args:
from tfx.extensions.google_cloud_ai_platform.pusher import executor as ai_platform_pusher_executor
pusher_kwargs.update({
"custom_executor_spec": executor_spec.ExecutorClassSpec(ai_platform_pusher_executor.Executor),
"custom_config": {
ai_platform_pusher_executor.SERVING_ARGS_KEY: ai_platform_serving_args}
})
elif serving_model_dir:
pusher_kwargs.update({
"push_destination": pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(base_directory=serving_model_dir))
})
else:
raise NotImplementedError("Provide ai_platform_serving_args or serving_model_dir.")
pusher = Pusher(**pusher_kwargs)
components = [
example_gen,
statistics_gen,
schema_gen,
example_validator,
transform,
trainer,
model_resolver,
evaluator,
pusher
]
return components
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
module_file: Text, serving_model_dir: Text,
metadata_path: Text,
direct_num_workers: int) -> pipeline.Pipeline:
"""Implements the chicago taxi pipeline with TFX."""
examples = external_input(data_root)
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input=examples)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
infer_schema = SchemaGen(
statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=False)
# Performs anomaly detection based on statistics and data schema.
validate_stats = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=infer_schema.outputs['schema'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(
examples=example_gen.outputs['examples'],
schema=infer_schema.outputs['schema'],
module_file=module_file)
# Uses user-provided Python function that implements a model using TF-Learn.
trainer = Trainer(
module_file=module_file,
transformed_examples=transform.outputs['transformed_examples'],
schema=infer_schema.outputs['schema'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=10000),
eval_args=trainer_pb2.EvalArgs(num_steps=5000))
# Uses TFMA to compute a evaluation statistics over features of a model.
model_analyzer = Evaluator(
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
feature_slicing_spec=evaluator_pb2.FeatureSlicingSpec(specs=[
evaluator_pb2.SingleSlicingSpec(
column_for_slicing=['trip_start_hour'])
]))
# Performs quality validation of a candidate model (compared to a baseline).
model_validator = ModelValidator(
examples=example_gen.outputs['examples'], model=trainer.outputs['model'])
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher = Pusher(
model=trainer.outputs['model'],
model_blessing=model_validator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen, statistics_gen, infer_schema, validate_stats, transform,
trainer, model_analyzer, model_validator, pusher
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
beam_pipeline_args=['--direct_num_workers=%d' % direct_num_workers])
from tfx.components import CsvExampleGen example_gen = CsvExampleGen(input_base="gs://example_compliance_data/")
def main(data_dir):
CsvExampleGen(input=external_input(str(data_dir)))
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
module_file: Text, serving_model_dir: Text,
beam_pipeline_args: List[Text]) -> pipeline.Pipeline:
"""Implements the chicago taxi pipeline with TFX and Kubeflow Pipelines."""
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input_base=data_root)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=False)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# Performs transformations and feature engineering in training and serving.
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=module_file)
# Uses user-provided Python function that implements a model using TF-Learn
# to train a model on Google Cloud AI Platform.
trainer = Trainer(
module_file=module_file,
transformed_examples=transform.outputs['transformed_examples'],
schema=schema_gen.outputs['schema'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=10000),
eval_args=trainer_pb2.EvalArgs(num_steps=5000),
)
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
# Uses TFMA to compute a evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(signature_name='eval')],
slicing_specs=[
tfma.SlicingSpec(),
tfma.SlicingSpec(feature_keys=['trip_start_hour'])
],
metrics_specs=[
tfma.MetricsSpec(
thresholds={
'accuracy':
tfma.config.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.6}),
# Change threshold will be ignored if there is no
# baseline model resolved from MLMD (first run).
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10}))
})
])
evaluator = Evaluator(examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
eval_config=eval_config)
# Performs infra validation of a candidate model to prevent unservable model
# from being pushed. In order to use InfraValidator component, persistent
# volume and its claim that the pipeline is using should be a ReadWriteMany
# access mode.
infra_validator = InfraValidator(
model=trainer.outputs['model'],
examples=example_gen.outputs['examples'],
serving_spec=infra_validator_pb2.ServingSpec(
tensorflow_serving=infra_validator_pb2.TensorFlowServing(
tags=['latest']),
kubernetes=infra_validator_pb2.KubernetesConfig()),
request_spec=infra_validator_pb2.RequestSpec(
tensorflow_serving=infra_validator_pb2.
TensorFlowServingRequestSpec()))
# Checks whether the model passed the validation steps and pushes the model
# to Google Cloud AI Platform if check passed.
pusher = Pusher(model=trainer.outputs['model'],
model_blessing=evaluator.outputs['blessing'],
infra_blessing=infra_validator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
return pipeline.Pipeline(pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen,
statistics_gen,
schema_gen,
example_validator,
transform,
trainer,
model_resolver,
evaluator,
infra_validator,
pusher,
],
beam_pipeline_args=beam_pipeline_args)
def _create_pipeline(pipeline_name: Text, pipeline_root: Text, data_root: Text,
module_file: Text, serving_model_dir: Text,
metadata_path: Text) -> pipeline.Pipeline:
"""Implements the Iris flowers pipeline with TFX."""
examples = external_input(data_root)
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input=examples)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=True)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# Hyperparameter tuning based on the tuner_fn in module_file.
tuner = Tuner(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=module_file)
# Uses user-provided Python function that implements a model using TF-Learn.
# TODO(jyzhao): example for importing a hyperparameters file generated not in
# currently run, e.g., by previous pipeline run with Tuner.
# TODO(jyzhao): consider supporting warmstart from tuner's model for trainer.
trainer = Trainer(module_file=module_file,
examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
hyperparameters=tuner.outputs['best_hyperparameters'],
train_args=trainer_pb2.TrainArgs(num_steps=10000),
eval_args=trainer_pb2.EvalArgs(num_steps=5000))
# Uses TFMA to compute a evaluation statistics over features of a model.
evaluator = Evaluator(examples=example_gen.outputs['examples'],
model=trainer.outputs['model'])
# Performs quality validation of a candidate model (compared to a baseline).
model_validator = ModelValidator(examples=example_gen.outputs['examples'],
model=trainer.outputs['model'])
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher = Pusher(model=trainer.outputs['model'],
model_blessing=model_validator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)))
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=[
example_gen,
statistics_gen,
schema_gen,
example_validator,
tuner,
trainer,
evaluator,
model_validator,
pusher,
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
metadata_path),
)
def testTaxiPipelineNewStyleCompatibility(self):
examples = external_input('/tmp/fake/path')
example_gen = CsvExampleGen(input=examples)
self.assertIs(example_gen.inputs['input'],
example_gen.inputs['input_base'])
statistics_gen = StatisticsGen(
examples=example_gen.outputs['examples'])
self.assertIs(statistics_gen.inputs['examples'],
statistics_gen.inputs['input_data'])
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'])
self.assertIs(schema_gen.inputs['statistics'],
schema_gen.inputs['stats'])
self.assertIs(schema_gen.outputs['schema'],
schema_gen.outputs['output'])
example_validator = ExampleValidator(
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
self.assertIs(example_validator.inputs['statistics'],
example_validator.inputs['stats'])
self.assertIs(example_validator.outputs['anomalies'],
example_validator.outputs['output'])
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file='/tmp/fake/module/file')
self.assertIs(transform.inputs['examples'],
transform.inputs['input_data'])
self.assertIs(transform.outputs['transform_graph'],
transform.outputs['transform_output'])
trainer = Trainer(
module_file='/tmp/fake/module/file',
transformed_examples=transform.outputs['transformed_examples'],
schema=schema_gen.outputs['schema'],
transform_graph=transform.outputs['transform_graph'],
train_args=trainer_pb2.TrainArgs(num_steps=10000),
eval_args=trainer_pb2.EvalArgs(num_steps=5000))
self.assertIs(trainer.inputs['transform_graph'],
trainer.inputs['transform_output'])
self.assertIs(trainer.outputs['model'], trainer.outputs['output'])
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(signature_name='eval')],
slicing_specs=[
tfma.SlicingSpec(),
tfma.SlicingSpec(feature_keys=['trip_start_hour'])
],
metrics_specs=[
tfma.MetricsSpec(
thresholds={
'accuracy':
tfma.config.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.6}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.
HIGHER_IS_BETTER,
absolute={'value': -1e-10}))
})
])
evaluator = Evaluator(examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
eval_config=eval_config)
self.assertIs(evaluator.inputs['model'],
evaluator.inputs['model_exports'])
self.assertIs(evaluator.outputs['evaluation'],
evaluator.outputs['output'])
pusher = Pusher(model=trainer.outputs['output'],
model_blessing=evaluator.outputs['blessing'],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory='/fake/serving/dir')))
self.assertIs(pusher.inputs['model'], pusher.inputs['model_export'])
self.assertIs(pusher.outputs['pushed_model'],
pusher.outputs['model_push'])
def create_pipeline(
pipeline_name: Text,
pipeline_root: Text,
data_path: Text,
# TODO(step 7): (Optional) Uncomment here to use BigQuery as a data source.
# query: Text,
preprocessing_fn: Text,
run_fn: Text,
train_args: trainer_pb2.TrainArgs,
eval_args: trainer_pb2.EvalArgs,
eval_accuracy_threshold: float,
serving_model_dir: Text,
metadata_connection_config: Optional[
metadata_store_pb2.ConnectionConfig] = None,
beam_pipeline_args: Optional[List[Text]] = None,
ai_platform_training_args: Optional[Dict[Text, Text]] = None,
ai_platform_serving_args: Optional[Dict[Text, Any]] = None,
) -> pipeline.Pipeline:
"""Implements the chicago taxi pipeline with TFX."""
components = []
# Brings data into the pipeline or otherwise joins/converts training data.
example_gen = CsvExampleGen(input=external_input(data_path))
# TODO(step 7): (Optional) Uncomment here to use BigQuery as a data source.
# example_gen = big_query_example_gen_component.BigQueryExampleGen(
# query=query)
components.append(example_gen)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
# TODO(step 5): Uncomment here to add StatisticsGen to the pipeline.
components.append(statistics_gen)
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=True)
# TODO(step 5): Uncomment here to add SchemaGen to the pipeline.
components.append(schema_gen)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator( # pylint: disable=unused-variable
statistics=statistics_gen.outputs['statistics'],
schema=schema_gen.outputs['schema'])
# TODO(step 5): Uncomment here to add ExampleValidator to the pipeline.
components.append(example_validator)
# Performs transformations and feature engineering in training and serving.
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
preprocessing_fn=preprocessing_fn)
# TODO(step 6): Uncomment here to add Transform to the pipeline.
components.append(transform)
# Uses user-provided Python function that implements a model using TF-Learn.
trainer_args = {
'run_fn':
run_fn,
'transformed_examples':
transform.outputs['transformed_examples'],
'schema':
schema_gen.outputs['schema'],
'transform_graph':
transform.outputs['transform_graph'],
'train_args':
train_args,
'eval_args':
eval_args,
'custom_executor_spec':
executor_spec.ExecutorClassSpec(trainer_executor.GenericExecutor),
}
if ai_platform_training_args is not None:
trainer_args.update({
'custom_executor_spec':
executor_spec.ExecutorClassSpec(
ai_platform_trainer_executor.GenericExecutor),
'custom_config': {
ai_platform_trainer_executor.TRAINING_ARGS_KEY:
ai_platform_training_args,
}
})
trainer = Trainer(**trainer_args)
# TODO(step 6): Uncomment here to add Trainer to the pipeline.
components.append(trainer)
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
instance_name='latest_blessed_model_resolver',
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing))
# TODO(step 6): Uncomment here to add ResolverNode to the pipeline.
components.append(model_resolver)
# Uses TFMA to compute a evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
eval_config = tfma.EvalConfig(
model_specs=[tfma.ModelSpec(label_key='big_tipper')],
slicing_specs=[tfma.SlicingSpec()],
metrics_specs=[
tfma.MetricsSpec(metrics=[
tfma.MetricConfig(
class_name='BinaryAccuracy',
threshold=tfma.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': eval_accuracy_threshold}),
change_threshold=tfma.GenericChangeThreshold(
direction=tfma.MetricDirection.HIGHER_IS_BETTER,
absolute={'value': -1e-10})))
])
])
evaluator = Evaluator(
examples=example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
# Change threshold will be ignored if there is no baseline (first run).
eval_config=eval_config)
# TODO(step 6): Uncomment here to add Evaluator to the pipeline.
components.append(evaluator)
# Checks whether the model passed the validation steps and pushes the model
# to a file destination if check passed.
pusher_args = {
'model':
trainer.outputs['model'],
'model_blessing':
evaluator.outputs['blessing'],
'push_destination':
pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=serving_model_dir)),
}
if ai_platform_serving_args is not None:
pusher_args.update({
'custom_executor_spec':
executor_spec.ExecutorClassSpec(
ai_platform_pusher_executor.Executor),
'custom_config': {
ai_platform_pusher_executor.SERVING_ARGS_KEY:
ai_platform_serving_args
},
})
pusher = Pusher(**pusher_args) # pylint: disable=unused-variable
# TODO(step 6): Uncomment here to add Pusher to the pipeline.
components.append(pusher)
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root,
components=components,
# Change this value to control caching of execution results. Default value
# is `False`.
# enable_cache=True,
metadata_connection_config=metadata_connection_config,
beam_pipeline_args=beam_pipeline_args,
)
def create_pipeline(
pipeline_name: Text,
data_path: Text,
outputs_path: Text,
output_model_path: Text,
train_args: trainer_pb2.TrainArgs,
eval_args: trainer_pb2.EvalArgs,
eval_accuracy_threshold: float,
metadata_connection_config: Optional[
metadata_store_pb2.ConnectionConfig] = None,
beam_pipeline_args: Optional[List[Text]] = None,
) -> Pipeline:
"""Implements the penguin pipeline with TFX."""
# TODO: remove this!
data_path = DATA_PATH
components = []
# Brings data into the pipeline or otherwise joins/converts training data.
# TODO(step 2): Might use another ExampleGen class for your data.
example_gen = CsvExampleGen(input=external_input(data_path))
components.append(example_gen)
# Computes statistics over data for visualization and example validation.
statistics_gen = StatisticsGen(examples=example_gen.outputs["examples"])
components.append(statistics_gen)
# Generates schema based on statistics files.
schema_gen = SchemaGen(statistics=statistics_gen.outputs["statistics"],
infer_feature_shape=True)
components.append(schema_gen)
# Performs anomaly detection based on statistics and data schema.
example_validator = ExampleValidator( # pylint: disable=unused-variable
statistics=statistics_gen.outputs["statistics"],
schema=schema_gen.outputs["schema"],
)
components.append(example_validator)
# Performs transformations and feature engineering in training and serving.
transform = Transform( # pylint: disable=unused-variable
examples=example_gen.outputs["examples"],
schema=schema_gen.outputs["schema"],
# TODO: we should be able to just pass: inputs={} and outputs={}
preprocessing_fn=f"{model.file_name}.preprocess",
)
# TODO(step 3): Uncomment here to add Transform to the pipeline.
components.append(transform)
# Uses user-provided Python function that implements a model using Tensorflow.
trainer = Trainer(
run_fn=f"{model.file_name}.train",
# run_fn="models.model.run_fn",
# examples=example_gen.outputs["examples"],
# Use outputs of Transform as training inputs if Transform is used.
examples=transform.outputs["transformed_examples"],
transform_graph=transform.outputs["transform_graph"],
schema=schema_gen.outputs["schema"],
train_args=train_args,
eval_args=eval_args,
custom_executor_spec=executor_spec.ExecutorClassSpec(
trainer_executor.GenericExecutor),
)
# TODO(step 4): Uncomment here to add Trainer to the pipeline.
components.append(trainer)
# Get the latest blessed model for model validation.
model_resolver = ResolverNode(
instance_name="latest_blessed_model_resolver",
resolver_class=latest_blessed_model_resolver.
LatestBlessedModelResolver,
model=Channel(type=Model),
model_blessing=Channel(type=ModelBlessing),
)
# TODO(step 5): Uncomment here to add ResolverNode to the pipeline.
components.append(model_resolver)
# Uses TFMA to compute a evaluation statistics over features of a model and
# perform quality validation of a candidate model (compared to a baseline).
evaluator = Evaluator( # pylint: disable=unused-variable
examples=example_gen.outputs["examples"],
model=trainer.outputs["model"],
baseline_model=model_resolver.outputs["model"],
# Change threshold will be ignored if there is no baseline (first run).
eval_config=tfm.Features(
model.outputs).eval_config(eval_accuracy_threshold),
)
# TODO(step 5): Uncomment here to add Evaluator to the pipeline.
components.append(evaluator)
# Pushes the model to a file destination if check passed.
pusher = Pusher( # pylint: disable=unused-variable
model=trainer.outputs["model"],
model_blessing=evaluator.outputs["blessing"],
push_destination=pusher_pb2.PushDestination(
filesystem=pusher_pb2.PushDestination.Filesystem(
base_directory=output_model_path)),
)
# TODO(step 5): Uncomment here to add Pusher to the pipeline.
components.append(pusher)
return Pipeline(
pipeline_name=pipeline_name,
pipeline_root=outputs_path,
components=components,
# Change this value to control caching of execution results. Default value
# is `False`.
# enable_cache=True,
metadata_connection_config=metadata_connection_config,
beam_pipeline_args=beam_pipeline_args,
)
