def testDoWithBlessedModel(self):
input_dict = {
'examples': [self._examples],
'model_export': [self._model_export],
'model_blessing': [self._model_blessing],
}
output_dict = {
'output': [self._inference_result],
}
# Create exe properties.
exec_properties = {
'data_spec':
json_format.MessageToJson(bulk_inferrer_pb2.DataSpec()),
'model_spec':
json_format.MessageToJson(bulk_inferrer_pb2.ModelSpec()),
'component_id': self.component_id,
}
# Run executor.
bulk_inferrer = executor.Executor(self._context)
bulk_inferrer.Do(input_dict, output_dict, exec_properties)
# Check outputs.
self.assertTrue(tf.io.gfile.exists(self._prediction_log_dir))
results = self._get_results(self._prediction_log_dir)
self.assertTrue(results)
self.assertEqual(
len(results[0].classify_log.response.result.classifications), 1)
self.assertEqual(
len(results[0].classify_log.response.result.classifications[0].
classes), 2)
def __init__(self,
examples: types.Channel = None,
model: Optional[types.Channel] = None,
model_blessing: Optional[types.Channel] = None,
data_spec: Optional[Union[bulk_inferrer_pb2.DataSpec,
Dict[Text, Any]]] = None,
model_spec: Optional[Union[bulk_inferrer_pb2.ModelSpec,
Dict[Text, Any]]] = None,
output_example_spec: Optional[Union[
bulk_inferrer_pb2.OutputExampleSpec, Dict[Text,
Any]]] = None):
"""Construct an BulkInferrer component.
Args:
examples: A Channel of type `standard_artifacts.Examples`, usually
produced by an ExampleGen component. _required_
model: A Channel of type `standard_artifacts.Model`, usually produced by
a Trainer component.
model_blessing: A Channel of type `standard_artifacts.ModelBlessing`,
usually produced by a ModelValidator component.
data_spec: bulk_inferrer_pb2.DataSpec instance that describes data
selection. If any field is provided as a RuntimeParameter, data_spec
should be constructed as a dict with the same field names as DataSpec
proto message.
model_spec: bulk_inferrer_pb2.ModelSpec instance that describes model
specification. If any field is provided as a RuntimeParameter,
model_spec should be constructed as a dict with the same field names as
ModelSpec proto message.
output_example_spec: bulk_inferrer_pb2.OutputExampleSpec instance, specify
if you want BulkInferrer to output examples instead of inference result.
If any field is provided as a RuntimeParameter, output_example_spec
should be constructed as a dict with the same field names as
OutputExampleSpec proto message.
"""
if output_example_spec:
output_examples = types.Channel(type=standard_artifacts.Examples)
inference_result = None
else:
inference_result = types.Channel(
type=standard_artifacts.InferenceResult)
output_examples = None
spec = BulkInferrerSpec(examples=examples,
model=model,
model_blessing=model_blessing,
data_spec=data_spec
or bulk_inferrer_pb2.DataSpec(),
model_spec=model_spec
or bulk_inferrer_pb2.ModelSpec(),
output_example_spec=output_example_spec,
inference_result=inference_result,
output_examples=output_examples)
super(BulkInferrer, self).__init__(spec=spec)
def _get_inference_spec(
self, model_path: Text,
exec_properties: Dict[Text, Any]) -> model_spec_pb2.InferenceSpecType:
model_spec = bulk_inferrer_pb2.ModelSpec()
proto_utils.json_to_proto(exec_properties['model_spec'], model_spec)
saved_model_spec = model_spec_pb2.SavedModelSpec(
model_path=model_path,
tag=model_spec.tag,
signature_name=model_spec.model_signature_name)
result = model_spec_pb2.InferenceSpecType()
result.saved_model_spec.CopyFrom(saved_model_spec)
return result
def _get_inference_spec(
self, model_path: str,
exec_properties: Dict[str, Any]) -> model_spec_pb2.InferenceSpecType:
model_spec = bulk_inferrer_pb2.ModelSpec()
proto_utils.json_to_proto(
exec_properties[standard_component_specs.MODEL_SPEC_KEY], model_spec)
saved_model_spec = model_spec_pb2.SavedModelSpec(
model_path=model_path,
tag=model_spec.tag,
signature_name=model_spec.model_signature_name)
result = model_spec_pb2.InferenceSpecType()
result.saved_model_spec.CopyFrom(saved_model_spec)
return result
def __init__(self,
examples: types.Channel = None,
model: Optional[types.Channel] = None,
model_blessing: Optional[types.Channel] = None,
data_spec: Optional[Union[bulk_inferrer_pb2.DataSpec,
Dict[Text, Any]]] = None,
model_spec: Optional[Union[bulk_inferrer_pb2.ModelSpec,
Dict[Text, Any]]] = None,
inference_result: Optional[types.Channel] = None,
instance_name: Optional[Text] = None,
enable_cache: Optional[bool] = None):
"""Construct an BulkInferrer component.
Args:
examples: A Channel of type `standard_artifacts.Examples`, usually
produced by an ExampleGen component. _required_
model: A Channel of type `standard_artifacts.Model`, usually produced by
a Trainer component.
model_blessing: A Channel of type `standard_artifacts.ModelBlessing`,
usually produced by a ModelValidator component.
data_spec: bulk_inferrer_pb2.DataSpec instance that describes data
selection. If any field is provided as a RuntimeParameter, data_spec
should be constructed as a dict with the same field names as DataSpec
proto message.
model_spec: bulk_inferrer_pb2.ModelSpec instance that describes model
specification. If any field is provided as a RuntimeParameter,
model_spec should be constructed as a dict with the same field names as
ModelSpec proto message.
inference_result: Channel of type `standard_artifacts.InferenceResult`
to store the inference results.
instance_name: Optional name assigned to this specific instance of
BulkInferrer. Required only if multiple BulkInferrer components are
declared in the same pipeline.
enable_cache: Optional boolean to indicate if cache is enabled for the
BulkInferrer component. If not specified, defaults to the value
specified for pipeline's enable_cache parameter.
"""
inference_result = inference_result or types.Channel(
type=standard_artifacts.InferenceResult,
artifacts=[standard_artifacts.InferenceResult()])
spec = BulkInferrerSpec(examples=examples,
model=model,
model_blessing=model_blessing,
data_spec=data_spec
or bulk_inferrer_pb2.DataSpec(),
model_spec=model_spec
or bulk_inferrer_pb2.ModelSpec(),
inference_result=inference_result)
super(BulkInferrer, self).__init__(spec=spec,
instance_name=instance_name,
enable_cache=enable_cache)
def __init__(
self,
examples: types.BaseChannel,
model: Optional[types.BaseChannel] = None,
model_blessing: Optional[types.BaseChannel] = None,
data_spec: Optional[Union[bulk_inferrer_pb2.DataSpec,
data_types.RuntimeParameter]] = None,
model_spec: Optional[Union[bulk_inferrer_pb2.ModelSpec,
data_types.RuntimeParameter]] = None,
output_example_spec: Optional[
Union[bulk_inferrer_pb2.OutputExampleSpec,
data_types.RuntimeParameter]] = None):
"""Construct an BulkInferrer component.
Args:
examples: A BaseChannel of type `standard_artifacts.Examples`, usually
produced by an ExampleGen component. _required_
model: A BaseChannel of type `standard_artifacts.Model`, usually produced
by a Trainer component.
model_blessing: A BaseChannel of type `standard_artifacts.ModelBlessing`,
usually produced by a ModelValidator component.
data_spec: bulk_inferrer_pb2.DataSpec instance that describes data
selection.
model_spec: bulk_inferrer_pb2.ModelSpec instance that describes model
specification.
output_example_spec: bulk_inferrer_pb2.OutputExampleSpec instance, specify
if you want BulkInferrer to output examples instead of inference result.
"""
if output_example_spec:
output_examples = types.Channel(type=standard_artifacts.Examples)
inference_result = None
else:
inference_result = types.Channel(
type=standard_artifacts.InferenceResult)
output_examples = None
spec = standard_component_specs.BulkInferrerSpec(
examples=examples,
model=model,
model_blessing=model_blessing,
data_spec=data_spec or bulk_inferrer_pb2.DataSpec(),
model_spec=model_spec or bulk_inferrer_pb2.ModelSpec(),
output_example_spec=output_example_spec,
inference_result=inference_result,
output_examples=output_examples)
super().__init__(spec=spec)
def __init__(self,
examples: types.Channel = None,
model_export: Optional[types.Channel] = None,
model_blessing: Optional[types.Channel] = None,
model_push: Optional[types.Channel] = None,
data_spec: Optional[bulk_inferrer_pb2.DataSpec] = None,
model_spec: Optional[bulk_inferrer_pb2.ModelSpec] = None,
output: Optional[types.Channel] = None,
instance_name: Optional[Text] = None):
"""Construct an BulkInferrer component.
Args:
examples: A Channel of 'ExamplesPath' type, usually produced by ExampleGen
component. _required_
model_export: A Channel of 'ModelExportPath' type, usually produced by
Trainer component.
model_blessing: A Channel of 'ModelBlessingPath' type, usually produced by
Model Validator component.
model_push: A Channel of 'PushedModel' type, usually produced by Pusher
component.
data_spec: bulk_inferrer_pb2.DataSpec instance that describes data
selection.
model_spec: bulk_inferrer_pb2.ModelSpec instance that describes model
specification.
output: Channel of `InferenceResult` to store the inference results.
instance_name: Optional name assigned to this specific instance of
BulkInferrer. Required only if multiple BulkInferrer components are
declared in the same pipeline.
"""
output = output or types.Channel(
type=standard_artifacts.InferenceResult,
artifacts=[standard_artifacts.InferenceResult()])
spec = BulkInferrerSpec(examples=examples,
model_export=model_export,
model_blessing=model_blessing,
model_push=model_push,
data_spec=data_spec
or bulk_inferrer_pb2.DataSpec(),
model_spec=model_spec
or bulk_inferrer_pb2.ModelSpec(),
output=output)
super(BulkInferrer, self).__init__(spec=spec,
instance_name=instance_name)
def generate_pipeline(pipeline_name, pipeline_root, train_data, test_data,
train_steps, eval_steps, pusher_target, runner):
module_file = 'util.py' # util.py is a file in the same folder
# RuntimeParameter is only supported on KubeflowDagRunner currently
if runner == 'kubeflow':
pipeline_root_param = os.path.join('gs://{{kfp-default-bucket}}',
pipeline_name, '{{workflow.uid}}')
train_data_param = data_types.RuntimeParameter(
name='train-data',
default=
'gs://renming-mlpipeline-kubeflowpipelines-default/kaggle/santander/train',
ptype=Text)
test_data_param = data_types.RuntimeParameter(
name='test-data',
default=
'gs://renming-mlpipeline-kubeflowpipelines-default/kaggle/santander/test',
ptype=Text)
pusher_target_param = data_types.RuntimeParameter(
name='pusher-destination',
default=
'gs://renming-mlpipeline-kubeflowpipelines-default/kaggle/santander/serving',
ptype=Text)
else:
pipeline_root_param = pipeline_root
train_data_param = train_data
test_data_param = test_data
pusher_target_param = pusher_target
examples = external_input(train_data_param)
example_gen = CsvExampleGen(input=examples, instance_name="train")
test_examples = external_input(test_data_param)
test_example_gen = CsvExampleGen(input=test_examples,
output_config={
'split_config': {
'splits': [{
'name': 'test',
'hash_buckets': 1
}]
}
},
instance_name="test")
statistics_gen = StatisticsGen(examples=example_gen.outputs['examples'])
schema_gen = SchemaGen(statistics=statistics_gen.outputs['statistics'],
infer_feature_shape=True
) # infer_feature_shape controls sparse or dense
# Transform is too slow in my side.
transform = Transform(examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=module_file)
trainer = Trainer(
custom_executor_spec=executor_spec.ExecutorClassSpec(GenericExecutor),
examples=transform.outputs['transformed_examples'],
transform_graph=transform.outputs['transform_graph'],
schema=schema_gen.outputs['schema'],
module_file=module_file,
train_args=trainer_pb2.TrainArgs(num_steps=train_steps),
eval_args=trainer_pb2.EvalArgs(num_steps=eval_steps),
instance_name="train",
enable_cache=False)
# 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(label_key='target')],
# tfma.SlicingSpec(feature_keys=['var_0', 'var_1']) when add more, Evaluator can't ouptput BLESSED status. It should be a bug in TFMA.
slicing_specs=[tfma.SlicingSpec()],
metrics_specs=[
tfma.MetricsSpec(
thresholds={
'binary_accuracy':
tfma.config.MetricThreshold(
value_threshold=tfma.GenericValueThreshold(
lower_bound={'value': 0.4}),
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,
instance_name="eval5")
# 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={
'filesystem': {
'base_directory': pusher_target_param
}
})
bulk_inferrer = BulkInferrer(
examples=test_example_gen.outputs['examples'],
model=trainer.outputs['model'],
# model_blessing=evaluator.outputs['blessing'],
data_spec=bulk_inferrer_pb2.DataSpec(),
model_spec=bulk_inferrer_pb2.ModelSpec(),
instance_name="bulkInferrer")
hello = component.HelloComponent(
input_data=bulk_inferrer.outputs['inference_result'],
instance_name='csvGen')
return pipeline.Pipeline(
pipeline_name=pipeline_name,
pipeline_root=pipeline_root_param,
components=[
example_gen, statistics_gen, schema_gen, transform, trainer,
model_resolver, evaluator, pusher, hello, test_example_gen,
bulk_inferrer
],
enable_cache=True,
metadata_connection_config=metadata.sqlite_metadata_connection_config(
os.path.join(pipeline_root, 'metadata.sqlite')),
beam_pipeline_args=['--direct_num_workers=0'])
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/141578059): 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,
training_data_root: Text, inference_data_root: Text,
module_file: Text, metadata_path: Text,
beam_pipeline_args: List[Text]) -> pipeline.Pipeline:
"""Implements the chicago taxi pipeline with TFX."""
# Brings training data into the pipeline or otherwise joins/converts
# training data.
training_example_gen = CsvExampleGen(input_base=training_data_root,
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 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=training_example_gen.outputs['examples'],
model=trainer.outputs['model'],
baseline_model=model_resolver.outputs['model'],
eval_config=eval_config)
# Brings inference data into the pipeline.
inference_example_gen = CsvExampleGen(
input_base=inference_data_root,
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),
beam_pipeline_args=beam_pipeline_args)
def Do(self, input_dict: Dict[Text, List[types.Artifact]],
output_dict: Dict[Text, List[types.Artifact]],
exec_properties: Dict[Text, Any]) -> None:
"""Runs batch inference on a given model with given input examples.
Args:
input_dict: Input dict from input key to a list of Artifacts.
- examples: examples for inference.
- model: exported model.
- model_blessing: model blessing result
output_dict: Output dict from output key to a list of Artifacts.
- output: bulk inference results.
exec_properties: A dict of execution properties.
- model_spec: JSON string of bulk_inferrer_pb2.ModelSpec instance.
- data_spec: JSON string of bulk_inferrer_pb2.DataSpec instance.
Returns:
None
"""
self._log_startup(input_dict, output_dict, exec_properties)
if 'examples' not in input_dict:
raise ValueError('\'examples\' is missing in input dict.')
if 'inference_result' not in output_dict:
raise ValueError('\'inference_result\' is missing in output dict.')
output = artifact_utils.get_single_instance(
output_dict['inference_result'])
if 'model' not in input_dict:
raise ValueError('Input models are not valid, model '
'need to be specified.')
if 'model_blessing' in input_dict:
model_blessing = artifact_utils.get_single_instance(
input_dict['model_blessing'])
if not model_utils.is_model_blessed(model_blessing):
output.set_int_custom_property('inferred', 0)
logging.info('Model on %s was not blessed', model_blessing.uri)
return
else:
logging.info(
'Model blessing is not provided, exported model will be '
'used.')
model = artifact_utils.get_single_instance(input_dict['model'])
model_path = path_utils.serving_model_path(model.uri)
logging.info('Use exported model from %s.', model_path)
data_spec = bulk_inferrer_pb2.DataSpec()
json_format.Parse(exec_properties['data_spec'], data_spec)
example_uris = {}
if data_spec.example_splits:
for example in input_dict['examples']:
for split in artifact_utils.decode_split_names(
example.split_names):
if split in data_spec.example_splits:
example_uris[split] = os.path.join(example.uri, split)
else:
for example in input_dict['examples']:
for split in artifact_utils.decode_split_names(
example.split_names):
example_uris[split] = os.path.join(example.uri, split)
model_spec = bulk_inferrer_pb2.ModelSpec()
json_format.Parse(exec_properties['model_spec'], model_spec)
output_path = os.path.join(output.uri, _PREDICTION_LOGS_DIR_NAME)
self._run_model_inference(model_path, example_uris, output_path,
model_spec)
logging.info('BulkInferrer generates prediction log to %s',
output_path)
output.set_int_custom_property('inferred', 1)
def setUp(self):
super(ExecutorTest, self).setUp()
self._source_data_dir = os.path.join(
os.path.dirname(os.path.dirname(__file__)), 'testdata')
self._output_data_dir = os.path.join(
os.environ.get('TEST_UNDECLARED_OUTPUTS_DIR', self.get_temp_dir()),
self._testMethodName)
self.component_id = 'test_component'
# Create input dict.
self._examples = standard_artifacts.Examples()
unlabelled_path = os.path.join(self._source_data_dir,
'csv_example_gen', 'unlabelled')
self._examples.uri = os.path.join(self._output_data_dir,
'csv_example_gen')
io_utils.copy_dir(unlabelled_path,
os.path.join(self._examples.uri, 'unlabelled'))
io_utils.copy_dir(unlabelled_path,
os.path.join(self._examples.uri, 'unlabelled2'))
self._examples.split_names = artifact_utils.encode_split_names(
['unlabelled', 'unlabelled2'])
self._model = standard_artifacts.Model()
self._model.uri = os.path.join(self._source_data_dir,
'trainer/current')
self._model_blessing = standard_artifacts.ModelBlessing()
self._model_blessing.uri = os.path.join(self._source_data_dir,
'model_validator/blessed')
self._model_blessing.set_int_custom_property('blessed', 1)
self._input_dict = {
'examples': [self._examples],
'model': [self._model],
'model_blessing': [self._model_blessing],
}
# Create output dict.
self._inference_result = standard_artifacts.InferenceResult()
self._prediction_log_dir = os.path.join(self._output_data_dir,
'prediction_logs')
self._inference_result.uri = self._prediction_log_dir
self._output_examples = standard_artifacts.Examples()
self._output_examples_dir = os.path.join(self._output_data_dir,
'output_examples')
self._output_examples.uri = self._output_examples_dir
self._output_dict_ir = {
'inference_result': [self._inference_result],
}
self._output_dict_oe = {
'output_examples': [self._output_examples],
}
# Create exe properties.
self._exec_properties = {
'data_spec':
proto_utils.proto_to_json(bulk_inferrer_pb2.DataSpec()),
'model_spec':
proto_utils.proto_to_json(bulk_inferrer_pb2.ModelSpec()),
'component_id': self.component_id,
}
# Create context
self._tmp_dir = os.path.join(self._output_data_dir, '.temp')
self._context = executor.Executor.Context(tmp_dir=self._tmp_dir,
unique_id='2')
def Do(self, input_dict: Dict[Text, List[types.Artifact]],
output_dict: Dict[Text, List[types.Artifact]],
exec_properties: Dict[Text, Any]) -> None:
"""Runs batch inference on a given model with given input examples.
Args:
input_dict: Input dict from input key to a list of Artifacts.
- examples: examples for inference.
- model: exported model.
- model_blessing: model blessing result, optional.
output_dict: Output dict from output key to a list of Artifacts.
- output: bulk inference results.
exec_properties: A dict of execution properties.
- model_spec: JSON string of bulk_inferrer_pb2.ModelSpec instance.
- data_spec: JSON string of bulk_inferrer_pb2.DataSpec instance.
Returns:
None
"""
self._log_startup(input_dict, output_dict, exec_properties)
source = exec_properties[StepKeys.SOURCE]
args = exec_properties[StepKeys.ARGS]
c = source_utils.load_source_path_class(source)
inferrer_step: BaseInferrer = c(**args)
output_examples = artifact_utils.get_single_instance(
output_dict[PREDICTIONS])
if EXAMPLES not in input_dict:
raise ValueError('\'examples\' is missing in input dict.')
if MODEL not in input_dict:
raise ValueError('Input models are not valid, model '
'need to be specified.')
if MODEL_BLESSING in input_dict:
model_blessing = artifact_utils.get_single_instance(
input_dict['model_blessing'])
if not model_utils.is_model_blessed(model_blessing):
logging.info('Model on %s was not blessed', model_blessing.uri)
return
else:
logging.info(
'Model blessing is not provided, exported model will be '
'used.')
model = artifact_utils.get_single_instance(input_dict[MODEL])
model_path = path_utils.serving_model_path(model.uri)
logging.info('Use exported model from %s.', model_path)
output_example_spec = bulk_inferrer_pb2.OutputExampleSpec(
output_columns_spec=[
bulk_inferrer_pb2.OutputColumnsSpec(
predict_output=bulk_inferrer_pb2.PredictOutput(
output_columns=[
bulk_inferrer_pb2.PredictOutputCol(
output_key=x,
output_column=f'{x}_label',
) for x in inferrer_step.get_labels()
]))
])
model_spec = bulk_inferrer_pb2.ModelSpec()
saved_model_spec = model_spec_pb2.SavedModelSpec(
model_path=model_path,
tag=model_spec.tag,
signature_name=model_spec.model_signature_name)
inference_spec = model_spec_pb2.InferenceSpecType()
inference_spec.saved_model_spec.CopyFrom(saved_model_spec)
self._run_model_inference(output_example_spec, input_dict[EXAMPLES],
output_examples, inference_spec,
inferrer_step)
def __init__(self,
examples: types.Channel = None,
model: Optional[types.Channel] = None,
model_blessing: Optional[types.Channel] = None,
data_spec: Optional[Union[bulk_inferrer_pb2.DataSpec,
Dict[Text, Any]]] = None,
model_spec: Optional[Union[bulk_inferrer_pb2.ModelSpec,
Dict[Text, Any]]] = None,
output_example_spec: Optional[Union[
bulk_inferrer_pb2.OutputExampleSpec, Dict[Text,
Any]]] = None,
inference_result: Optional[types.Channel] = None,
output_examples: Optional[types.Channel] = None,
instance_name: Optional[Text] = None):
"""Construct an BulkInferrer component.
Args:
examples: A Channel of type `standard_artifacts.Examples`, usually
produced by an ExampleGen component. _required_
model: A Channel of type `standard_artifacts.Model`, usually produced by
a Trainer component.
model_blessing: A Channel of type `standard_artifacts.ModelBlessing`,
usually produced by a ModelValidator component.
data_spec: bulk_inferrer_pb2.DataSpec instance that describes data
selection. If any field is provided as a RuntimeParameter, data_spec
should be constructed as a dict with the same field names as DataSpec
proto message.
model_spec: bulk_inferrer_pb2.ModelSpec instance that describes model
specification. If any field is provided as a RuntimeParameter,
model_spec should be constructed as a dict with the same field names as
ModelSpec proto message.
output_example_spec: bulk_inferrer_pb2.OutputExampleSpec instance, specify
if you want BulkInferrer to output examples instead of inference result.
If any field is provided as a RuntimeParameter, output_example_spec
should be constructed as a dict with the same field names as
OutputExampleSpec proto message.
inference_result: Channel of type `standard_artifacts.InferenceResult`
to store the inference results, must not be specified when
output_example_spec is set.
output_examples: Channel of type `standard_artifacts.Examples`
to store the output examples, must not be specified when
output_example_spec is unset. Check output_example_spec for details.
instance_name: Optional name assigned to this specific instance of
BulkInferrer. Required only if multiple BulkInferrer components are
declared in the same pipeline.
Raises:
ValueError: Must not specify inference_result or output_examples depends
on whether output_example_spec is set or not.
"""
if output_example_spec:
if inference_result:
raise ValueError(
'Must not specify inference_result when output_example_spec is set.'
)
output_examples = output_examples or types.Channel(
type=standard_artifacts.Examples)
else:
if output_examples:
raise ValueError(
'Must not specify output_examples when output_example_spec is unset.'
)
inference_result = inference_result or types.Channel(
type=standard_artifacts.InferenceResult)
spec = BulkInferrerSpec(examples=examples,
model=model,
model_blessing=model_blessing,
data_spec=data_spec
or bulk_inferrer_pb2.DataSpec(),
model_spec=model_spec
or bulk_inferrer_pb2.ModelSpec(),
output_example_spec=output_example_spec,
inference_result=inference_result,
output_examples=output_examples)
super(BulkInferrer, self).__init__(spec=spec,
instance_name=instance_name)
def Do(self, input_dict: Dict[Text, List[types.Artifact]],
output_dict: Dict[Text, List[types.Artifact]],
exec_properties: Dict[Text, Any]) -> None:
"""Runs batch inference on a given model with given input examples.
Args:
input_dict: Input dict from input key to a list of Artifacts.
- examples: examples for inference.
- model_export: exported model.
- model_blessing: model blessing result
- model_push: pushed model Either model_push or (model_export and
model_blessing) need to present.
output_dict: Output dict from output key to a list of Artifacts.
- output: bulk inference results.
exec_properties: A dict of execution properties.
- model_spec: JSON string of bulk_inferrer_pb2.ModelSpec instance.
- data_spec: JSON string of bulk_inferrer_pb2.DataSpec instance.
Returns:
None
"""
self._log_startup(input_dict, output_dict, exec_properties)
if 'examples' not in input_dict:
raise ValueError('\'examples\' is missing in input dict.')
if 'output' not in output_dict:
raise ValueError('\'output\' is missing in output dict.')
output = artifact_utils.get_single_instance(output_dict['output'])
if 'model_push' in input_dict:
model_push = artifact_utils.get_single_instance(
input_dict['model_push'])
model_path = io_utils.get_only_uri_in_dir(model_push.uri)
logging.info('Use pushed model from %s.', model_path)
elif 'model_blessing' in input_dict and 'model_export' in input_dict:
model_blessing = artifact_utils.get_single_instance(
input_dict['model_blessing'])
if not model_utils.is_model_blessed(model_blessing):
output.set_int_custom_property('inferred', 0)
logging.info('Model on %s was not blessed', model_blessing.uri)
return
model_export = artifact_utils.get_single_instance(
input_dict['model_export'])
model_path = path_utils.serving_model_path(model_export.uri)
logging.info('Use exported model from %s.', model_path)
else:
raise ValueError(
'Input models are not valid. Either model_push or '
'(model_blessing and model_export) need to be '
'specified.')
data_spec = bulk_inferrer_pb2.DataSpec()
json_format.Parse(exec_properties['data_spec'], data_spec)
example_uris = {}
if data_spec.example_splits:
for example in input_dict['examples']:
if example.split in data_spec.example_splits:
example_uris[example.split] = example.uri
else:
for example in input_dict['examples']:
example_uris[example.split] = example.uri
model_spec = bulk_inferrer_pb2.ModelSpec()
json_format.Parse(exec_properties['model_spec'], model_spec)
output_path = os.path.join(output.uri, _PREDICTION_LOGS_DIR_NAME)
self._run_model_inference(model_path, example_uris, output_path,
model_spec)
logging.info('BulkInferrer generates prediction log to %s',
output_path)
output.set_int_custom_property('inferred', 1)