Steps
zenml.steps
special
Initializer for ZenML steps.
A step is a single piece or stage of a ZenML pipeline. Think of each step as being one of the nodes of a Directed Acyclic Graph (or DAG). Steps are responsible for one aspect of processing or interacting with the data / artifacts in the pipeline.
Conceptually, a Step is a discrete and independent part of a pipeline that is responsible for one particular aspect of data manipulation inside a ZenML pipeline.
Steps can be subclassed from the BaseStep
class, or used via our @step
decorator.
base_parameters
Step parameters.
BaseParameters (BaseModel)
Base class to pass parameters into a step.
Source code in zenml/steps/base_parameters.py
class BaseParameters(BaseModel):
"""Base class to pass parameters into a step."""
base_step
Base Step for ZenML.
BaseStep
Abstract base class for all ZenML steps.
Source code in zenml/steps/base_step.py
class BaseStep(metaclass=BaseStepMeta):
"""Abstract base class for all ZenML steps."""
def __init__(
self,
*args: Any,
name: Optional[str] = None,
enable_cache: Optional[bool] = None,
enable_artifact_metadata: Optional[bool] = None,
enable_artifact_visualization: Optional[bool] = None,
enable_step_logs: Optional[bool] = None,
experiment_tracker: Optional[str] = None,
step_operator: Optional[str] = None,
parameters: Optional["ParametersOrDict"] = None,
output_materializers: Optional[
"OutputMaterializersSpecification"
] = None,
settings: Optional[Mapping[str, "SettingsOrDict"]] = None,
extra: Optional[Dict[str, Any]] = None,
on_failure: Optional["HookSpecification"] = None,
on_success: Optional["HookSpecification"] = None,
model: Optional["Model"] = None,
retry: Optional[StepRetryConfig] = None,
**kwargs: Any,
) -> None:
"""Initializes a step.
Args:
*args: Positional arguments passed to the step.
name: The name of the step.
enable_cache: If caching should be enabled for this step.
enable_artifact_metadata: If artifact metadata should be enabled
for this step.
enable_artifact_visualization: If artifact visualization should be
enabled for this step.
enable_step_logs: Enable step logs for this step.
experiment_tracker: The experiment tracker to use for this step.
step_operator: The step operator to use for this step.
parameters: Function parameters for this step
output_materializers: Output materializers for this step. If
given as a dict, the keys must be a subset of the output names
of this step. If a single value (type or string) is given, the
materializer will be used for all outputs.
settings: settings for this step.
extra: Extra configurations for this step.
on_failure: Callback function in event of failure of the step. Can
be a function with a single argument of type `BaseException`, or
a source path to such a function (e.g. `module.my_function`).
on_success: Callback function in event of success of the step. Can
be a function with no arguments, or a source path to such a
function (e.g. `module.my_function`).
model: configuration of the model version in the Model Control Plane.
retry: Configuration for retrying the step in case of failure.
**kwargs: Keyword arguments passed to the step.
"""
from zenml.config.step_configurations import PartialStepConfiguration
self._upstream_steps: Set["BaseStep"] = set()
self.entrypoint_definition = validate_entrypoint_function(
self.entrypoint, reserved_arguments=["after", "id"]
)
name = name or self.__class__.__name__
requires_context = self.entrypoint_definition.context is not None
if enable_cache is None:
if requires_context:
# Using the StepContext inside a step provides access to
# external resources which might influence the step execution.
# We therefore disable caching unless it is explicitly enabled
enable_cache = False
logger.debug(
"Step `%s`: Step context required and caching not "
"explicitly enabled.",
name,
)
logger.debug(
"Step `%s`: Caching %s.",
name,
"enabled" if enable_cache is not False else "disabled",
)
logger.debug(
"Step `%s`: Artifact metadata %s.",
name,
"enabled" if enable_artifact_metadata is not False else "disabled",
)
logger.debug(
"Step `%s`: Artifact visualization %s.",
name,
"enabled"
if enable_artifact_visualization is not False
else "disabled",
)
logger.debug(
"Step `%s`: logs %s.",
name,
"enabled" if enable_step_logs is not False else "disabled",
)
if model is not None:
logger.debug(
"Step `%s`: Is in Model context %s.",
name,
{
"model": model.name,
"version": model.version,
},
)
self._configuration = PartialStepConfiguration(
name=name,
enable_cache=enable_cache,
enable_artifact_metadata=enable_artifact_metadata,
enable_artifact_visualization=enable_artifact_visualization,
enable_step_logs=enable_step_logs,
)
self.configure(
experiment_tracker=experiment_tracker,
step_operator=step_operator,
output_materializers=output_materializers,
parameters=parameters,
settings=settings,
extra=extra,
on_failure=on_failure,
on_success=on_success,
model=model,
retry=retry,
)
self._verify_and_apply_init_params(*args, **kwargs)
notebook_utils.try_to_save_notebook_cell_code(self.source_object)
@abstractmethod
def entrypoint(self, *args: Any, **kwargs: Any) -> Any:
"""Abstract method for core step logic.
Args:
*args: Positional arguments passed to the step.
**kwargs: Keyword arguments passed to the step.
Returns:
The output of the step.
"""
@classmethod
def load_from_source(cls, source: Union[Source, str]) -> "BaseStep":
"""Loads a step from source.
Args:
source: The path to the step source.
Returns:
The loaded step.
Raises:
ValueError: If the source is not a valid step source.
"""
obj = source_utils.load(source)
if isinstance(obj, BaseStep):
return obj
elif isinstance(obj, type) and issubclass(obj, BaseStep):
return obj()
else:
raise ValueError("Invalid step source.")
def resolve(self) -> Source:
"""Resolves the step.
Returns:
The step source.
"""
return source_utils.resolve(self.__class__)
@property
def upstream_steps(self) -> Set["BaseStep"]:
"""Names of the upstream steps of this step.
This property will only contain the full set of upstream steps once
it's parent pipeline `connect(...)` method was called.
Returns:
Set of upstream step names.
"""
return self._upstream_steps
def after(self, step: "BaseStep") -> None:
"""Adds an upstream step to this step.
Calling this method makes sure this step only starts running once the
given step has successfully finished executing.
**Note**: This can only be called inside the pipeline connect function
which is decorated with the `@pipeline` decorator. Any calls outside
this function will be ignored.
Example:
The following pipeline will run its steps sequentially in the following
order: step_2 -> step_1 -> step_3
```python
@pipeline
def example_pipeline(step_1, step_2, step_3):
step_1.after(step_2)
step_3(step_1(), step_2())
```
Args:
step: A step which should finish executing before this step is
started.
"""
self._upstream_steps.add(step)
@property
def source_object(self) -> Any:
"""The source object of this step.
Returns:
The source object of this step.
"""
return self.__class__
@property
def source_code(self) -> str:
"""The source code of this step.
Returns:
The source code of this step.
"""
return inspect.getsource(self.source_object)
@property
def docstring(self) -> Optional[str]:
"""The docstring of this step.
Returns:
The docstring of this step.
"""
return self.__doc__
@property
def caching_parameters(self) -> Dict[str, Any]:
"""Caching parameters for this step.
Returns:
A dictionary containing the caching parameters
"""
parameters = {
STEP_SOURCE_PARAMETER_NAME: source_code_utils.get_hashed_source_code(
self.source_object
)
}
for name, output in self.configuration.outputs.items():
if output.materializer_source:
key = f"{name}_materializer_source"
hash_ = hashlib.md5() # nosec
for source in output.materializer_source:
materializer_class = source_utils.load(source)
code_hash = source_code_utils.get_hashed_source_code(
materializer_class
)
hash_.update(code_hash.encode())
parameters[key] = hash_.hexdigest()
return parameters
def _verify_and_apply_init_params(self, *args: Any, **kwargs: Any) -> None:
"""Verifies the initialization args and kwargs of this step.
This method makes sure that there is only one parameters object passed
at initialization and that it was passed using the correct name and
type specified in the step declaration.
Args:
*args: The args passed to the init method of this step.
**kwargs: The kwargs passed to the init method of this step.
Raises:
StepInterfaceError: If there are too many arguments or arguments
with a wrong name/type.
"""
maximum_arg_count = (
1 if self.entrypoint_definition.legacy_params else 0
)
arg_count = len(args) + len(kwargs)
if arg_count > maximum_arg_count:
raise StepInterfaceError(
f"Too many arguments ({arg_count}, expected: "
f"{maximum_arg_count}) passed when creating a "
f"'{self.name}' step."
)
if self.entrypoint_definition.legacy_params:
if args:
config = args[0]
elif kwargs:
key, config = kwargs.popitem()
if key != self.entrypoint_definition.legacy_params.name:
raise StepInterfaceError(
f"Unknown keyword argument '{key}' when creating a "
f"'{self.name}' step, only expected a single "
"argument with key "
f"'{self.entrypoint_definition.legacy_params.name}'."
)
else:
# This step requires configuration parameters but no parameters
# object was passed as an argument. The parameters might be
# set via default values in the parameters class or in a
# configuration file, so we continue for now and verify
# that all parameters are set before running the step
return
if not isinstance(
config, self.entrypoint_definition.legacy_params.annotation
):
raise StepInterfaceError(
f"`{config}` object passed when creating a "
f"'{self.name}' step is not a "
f"`{self.entrypoint_definition.legacy_params.annotation.__name__} "
"` instance."
)
self.configure(parameters=config)
def _parse_call_args(
self, *args: Any, **kwargs: Any
) -> Tuple[
Dict[str, "StepArtifact"],
Dict[str, "ExternalArtifact"],
Dict[str, "ModelVersionDataLazyLoader"],
Dict[str, "ClientLazyLoader"],
Dict[str, Any],
Dict[str, Any],
]:
"""Parses the call args for the step entrypoint.
Args:
*args: Entrypoint function arguments.
**kwargs: Entrypoint function keyword arguments.
Raises:
StepInterfaceError: If invalid function arguments were passed.
Returns:
The artifacts, external artifacts, model version artifacts/metadata and parameters for the step.
"""
from zenml.artifacts.external_artifact import ExternalArtifact
from zenml.model.lazy_load import ModelVersionDataLazyLoader
from zenml.models.v2.core.artifact_version import (
LazyArtifactVersionResponse,
)
from zenml.models.v2.core.run_metadata import LazyRunMetadataResponse
signature = get_step_entrypoint_signature(step=self)
try:
bound_args = signature.bind_partial(*args, **kwargs)
except TypeError as e:
raise StepInterfaceError(
f"Wrong arguments when calling step '{self.name}': {e}"
) from e
artifacts = {}
external_artifacts = {}
model_artifacts_or_metadata = {}
client_lazy_loaders = {}
parameters = {}
default_parameters = {}
for key, value in bound_args.arguments.items():
self.entrypoint_definition.validate_input(key=key, value=value)
if isinstance(value, StepArtifact):
artifacts[key] = value
if key in self.configuration.parameters:
logger.warning(
"Got duplicate value for step input %s, using value "
"provided as artifact.",
key,
)
elif isinstance(value, ExternalArtifact):
external_artifacts[key] = value
if not value.id:
# If the external artifact references a fixed artifact by
# ID, caching behaves as expected.
logger.warning(
"Using an external artifact as step input currently "
"invalidates caching for the step and all downstream "
"steps. Future releases will introduce hashing of "
"artifacts which will improve this behavior."
)
elif isinstance(value, LazyArtifactVersionResponse):
model_artifacts_or_metadata[key] = ModelVersionDataLazyLoader(
model=value.lazy_load_model,
artifact_name=value.lazy_load_name,
artifact_version=value.lazy_load_version,
metadata_name=None,
)
elif isinstance(value, LazyRunMetadataResponse):
model_artifacts_or_metadata[key] = ModelVersionDataLazyLoader(
model=value.lazy_load_model,
artifact_name=value.lazy_load_artifact_name,
artifact_version=value.lazy_load_artifact_version,
metadata_name=value.lazy_load_metadata_name,
)
elif isinstance(value, ClientLazyLoader):
client_lazy_loaders[key] = value
else:
parameters[key] = value
# Above we iterated over the provided arguments which should overwrite
# any parameters previously defined on the step instance. Now we apply
# the default values on the entrypoint function and add those as
# parameters for any argument that has no value yet. If we were to do
# that in the above loop, we would overwrite previously configured
# parameters with the default values.
bound_args.apply_defaults()
for key, value in bound_args.arguments.items():
self.entrypoint_definition.validate_input(key=key, value=value)
if (
key not in artifacts
and key not in external_artifacts
and key not in model_artifacts_or_metadata
and key not in self.configuration.parameters
and key not in client_lazy_loaders
):
default_parameters[key] = value
return (
artifacts,
external_artifacts,
model_artifacts_or_metadata,
client_lazy_loaders,
parameters,
default_parameters,
)
def __call__(
self,
*args: Any,
id: Optional[str] = None,
after: Union[str, Sequence[str], None] = None,
**kwargs: Any,
) -> Any:
"""Handle a call of the step.
This method does one of two things:
* If there is an active pipeline context, it adds an invocation of the
step instance to the pipeline.
* If no pipeline is active, it calls the step entrypoint function.
Args:
*args: Entrypoint function arguments.
id: Invocation ID to use.
after: Upstream steps for the invocation.
**kwargs: Entrypoint function keyword arguments.
Returns:
The outputs of the entrypoint function call.
"""
from zenml.new.pipelines.pipeline import Pipeline
if not Pipeline.ACTIVE_PIPELINE:
# The step is being called outside the context of a pipeline, either
# run the step function or run it as a single step pipeline on the
# active stack
run_without_stack = handle_bool_env_var(
ENV_ZENML_RUN_SINGLE_STEPS_WITHOUT_STACK, default=False
)
if run_without_stack:
return self.call_entrypoint(*args, **kwargs)
else:
return run_as_single_step_pipeline(self, *args, **kwargs)
(
input_artifacts,
external_artifacts,
model_artifacts_or_metadata,
client_lazy_loaders,
parameters,
default_parameters,
) = self._parse_call_args(*args, **kwargs)
upstream_steps = {
artifact.invocation_id for artifact in input_artifacts.values()
}
if isinstance(after, str):
upstream_steps.add(after)
elif isinstance(after, Sequence):
upstream_steps = upstream_steps.union(after)
invocation_id = Pipeline.ACTIVE_PIPELINE.add_step_invocation(
step=self,
input_artifacts=input_artifacts,
external_artifacts=external_artifacts,
model_artifacts_or_metadata=model_artifacts_or_metadata,
client_lazy_loaders=client_lazy_loaders,
parameters=parameters,
default_parameters=default_parameters,
upstream_steps=upstream_steps,
custom_id=id,
allow_id_suffix=not id,
)
outputs = []
for key, annotation in self.entrypoint_definition.outputs.items():
output = StepArtifact(
invocation_id=invocation_id,
output_name=key,
annotation=annotation,
pipeline=Pipeline.ACTIVE_PIPELINE,
)
outputs.append(output)
return outputs[0] if len(outputs) == 1 else outputs
def call_entrypoint(self, *args: Any, **kwargs: Any) -> Any:
"""Calls the entrypoint function of the step.
Args:
*args: Entrypoint function arguments.
**kwargs: Entrypoint function keyword arguments.
Returns:
The return value of the entrypoint function.
Raises:
StepInterfaceError: If the arguments to the entrypoint function are
invalid.
"""
try:
validated_args = pydantic_utils.validate_function_args(
self.entrypoint,
ConfigDict(arbitrary_types_allowed=True),
*args,
**kwargs,
)
except ValidationError as e:
raise StepInterfaceError(
"Invalid step function entrypoint arguments. Check out the "
"pydantic error above for more details."
) from e
return self.entrypoint(**validated_args)
@property
def name(self) -> str:
"""The name of the step.
Returns:
The name of the step.
"""
return self.configuration.name
@property
def enable_cache(self) -> Optional[bool]:
"""If caching is enabled for the step.
Returns:
If caching is enabled for the step.
"""
return self.configuration.enable_cache
@property
def configuration(self) -> "PartialStepConfiguration":
"""The configuration of the step.
Returns:
The configuration of the step.
"""
return self._configuration
def configure(
self: T,
name: Optional[str] = None,
enable_cache: Optional[bool] = None,
enable_artifact_metadata: Optional[bool] = None,
enable_artifact_visualization: Optional[bool] = None,
enable_step_logs: Optional[bool] = None,
experiment_tracker: Optional[str] = None,
step_operator: Optional[str] = None,
parameters: Optional["ParametersOrDict"] = None,
output_materializers: Optional[
"OutputMaterializersSpecification"
] = None,
settings: Optional[Mapping[str, "SettingsOrDict"]] = None,
extra: Optional[Dict[str, Any]] = None,
on_failure: Optional["HookSpecification"] = None,
on_success: Optional["HookSpecification"] = None,
model: Optional["Model"] = None,
merge: bool = True,
retry: Optional[StepRetryConfig] = None,
) -> T:
"""Configures the step.
Configuration merging example:
* `merge==True`:
step.configure(extra={"key1": 1})
step.configure(extra={"key2": 2}, merge=True)
step.configuration.extra # {"key1": 1, "key2": 2}
* `merge==False`:
step.configure(extra={"key1": 1})
step.configure(extra={"key2": 2}, merge=False)
step.configuration.extra # {"key2": 2}
Args:
name: DEPRECATED: The name of the step.
enable_cache: If caching should be enabled for this step.
enable_artifact_metadata: If artifact metadata should be enabled
for this step.
enable_artifact_visualization: If artifact visualization should be
enabled for this step.
enable_step_logs: If step logs should be enabled for this step.
experiment_tracker: The experiment tracker to use for this step.
step_operator: The step operator to use for this step.
parameters: Function parameters for this step
output_materializers: Output materializers for this step. If
given as a dict, the keys must be a subset of the output names
of this step. If a single value (type or string) is given, the
materializer will be used for all outputs.
settings: settings for this step.
extra: Extra configurations for this step.
on_failure: Callback function in event of failure of the step. Can
be a function with a single argument of type `BaseException`, or
a source path to such a function (e.g. `module.my_function`).
on_success: Callback function in event of success of the step. Can
be a function with no arguments, or a source path to such a
function (e.g. `module.my_function`).
model: configuration of the model version in the Model Control Plane.
merge: If `True`, will merge the given dictionary configurations
like `parameters` and `settings` with existing
configurations. If `False` the given configurations will
overwrite all existing ones. See the general description of this
method for an example.
retry: Configuration for retrying the step in case of failure.
Returns:
The step instance that this method was called on.
"""
from zenml.config.step_configurations import StepConfigurationUpdate
from zenml.hooks.hook_validators import resolve_and_validate_hook
if name:
logger.warning("Configuring the name of a step is deprecated.")
def _resolve_if_necessary(
value: Union[str, Source, Type[Any]],
) -> Source:
if isinstance(value, str):
return Source.from_import_path(value)
elif isinstance(value, Source):
return value
else:
return source_utils.resolve(value)
def _convert_to_tuple(value: Any) -> Tuple[Source, ...]:
if isinstance(value, str) or not isinstance(value, Sequence):
return (_resolve_if_necessary(value),)
else:
return tuple(_resolve_if_necessary(v) for v in value)
outputs: Dict[str, Dict[str, Tuple[Source, ...]]] = defaultdict(dict)
allowed_output_names = set(self.entrypoint_definition.outputs)
if output_materializers:
if not isinstance(output_materializers, Mapping):
sources = _convert_to_tuple(output_materializers)
output_materializers = {
output_name: sources
for output_name in allowed_output_names
}
for output_name, materializer in output_materializers.items():
sources = _convert_to_tuple(materializer)
outputs[output_name]["materializer_source"] = sources
failure_hook_source = None
if on_failure:
# string of on_failure hook function to be used for this step
failure_hook_source = resolve_and_validate_hook(on_failure)
success_hook_source = None
if on_success:
# string of on_success hook function to be used for this step
success_hook_source = resolve_and_validate_hook(on_success)
if isinstance(parameters, BaseParameters):
parameters = parameters.model_dump()
values = dict_utils.remove_none_values(
{
"enable_cache": enable_cache,
"enable_artifact_metadata": enable_artifact_metadata,
"enable_artifact_visualization": enable_artifact_visualization,
"enable_step_logs": enable_step_logs,
"experiment_tracker": experiment_tracker,
"step_operator": step_operator,
"parameters": parameters,
"settings": settings,
"outputs": outputs or None,
"extra": extra,
"failure_hook_source": failure_hook_source,
"success_hook_source": success_hook_source,
"model": model,
"retry": retry,
}
)
config = StepConfigurationUpdate(**values)
self._apply_configuration(config, merge=merge)
return self
def with_options(
self,
enable_cache: Optional[bool] = None,
enable_artifact_metadata: Optional[bool] = None,
enable_artifact_visualization: Optional[bool] = None,
enable_step_logs: Optional[bool] = None,
experiment_tracker: Optional[str] = None,
step_operator: Optional[str] = None,
parameters: Optional["ParametersOrDict"] = None,
output_materializers: Optional[
"OutputMaterializersSpecification"
] = None,
settings: Optional[Mapping[str, "SettingsOrDict"]] = None,
extra: Optional[Dict[str, Any]] = None,
on_failure: Optional["HookSpecification"] = None,
on_success: Optional["HookSpecification"] = None,
model: Optional["Model"] = None,
merge: bool = True,
) -> "BaseStep":
"""Copies the step and applies the given configurations.
Args:
enable_cache: If caching should be enabled for this step.
enable_artifact_metadata: If artifact metadata should be enabled
for this step.
enable_artifact_visualization: If artifact visualization should be
enabled for this step.
enable_step_logs: If step logs should be enabled for this step.
experiment_tracker: The experiment tracker to use for this step.
step_operator: The step operator to use for this step.
parameters: Function parameters for this step
output_materializers: Output materializers for this step. If
given as a dict, the keys must be a subset of the output names
of this step. If a single value (type or string) is given, the
materializer will be used for all outputs.
settings: settings for this step.
extra: Extra configurations for this step.
on_failure: Callback function in event of failure of the step. Can
be a function with a single argument of type `BaseException`, or
a source path to such a function (e.g. `module.my_function`).
on_success: Callback function in event of success of the step. Can
be a function with no arguments, or a source path to such a
function (e.g. `module.my_function`).
model: configuration of the model version in the Model Control Plane.
merge: If `True`, will merge the given dictionary configurations
like `parameters` and `settings` with existing
configurations. If `False` the given configurations will
overwrite all existing ones. See the general description of this
method for an example.
Returns:
The copied step instance.
"""
step_copy = self.copy()
step_copy.configure(
enable_cache=enable_cache,
enable_artifact_metadata=enable_artifact_metadata,
enable_artifact_visualization=enable_artifact_visualization,
enable_step_logs=enable_step_logs,
experiment_tracker=experiment_tracker,
step_operator=step_operator,
parameters=parameters,
output_materializers=output_materializers,
settings=settings,
extra=extra,
on_failure=on_failure,
on_success=on_success,
model=model,
merge=merge,
)
return step_copy
def copy(self) -> "BaseStep":
"""Copies the step.
Returns:
The step copy.
"""
return copy.deepcopy(self)
def _apply_configuration(
self,
config: "StepConfigurationUpdate",
merge: bool = True,
runtime_parameters: Dict[str, Any] = {},
) -> None:
"""Applies an update to the step configuration.
Args:
config: The configuration update.
runtime_parameters: Dictionary of parameters passed to a step from runtime
merge: Whether to merge the updates with the existing configuration
or not. See the `BaseStep.configure(...)` method for a detailed
explanation.
"""
self._validate_configuration(config, runtime_parameters)
self._configuration = pydantic_utils.update_model(
self._configuration, update=config, recursive=merge
)
logger.debug("Updated step configuration:")
logger.debug(self._configuration)
def _validate_configuration(
self,
config: "StepConfigurationUpdate",
runtime_parameters: Dict[str, Any],
) -> None:
"""Validates a configuration update.
Args:
config: The configuration update to validate.
runtime_parameters: Dictionary of parameters passed to a step from runtime
"""
settings_utils.validate_setting_keys(list(config.settings))
self._validate_function_parameters(
parameters=config.parameters, runtime_parameters=runtime_parameters
)
self._validate_outputs(outputs=config.outputs)
def _validate_function_parameters(
self,
parameters: Dict[str, Any],
runtime_parameters: Dict[str, Any],
) -> None:
"""Validates step function parameters.
Args:
parameters: The parameters to validate.
runtime_parameters: Dictionary of parameters passed to a step from runtime
Raises:
StepInterfaceError: If the step requires no function parameters but
parameters were configured.
RuntimeError: If the step has parameters configured differently in
configuration file and code.
"""
if not parameters:
return
conflicting_parameters = {}
for key, value in parameters.items():
if key in runtime_parameters:
runtime_value = runtime_parameters[key]
if runtime_value != value:
conflicting_parameters[key] = (value, runtime_value)
if key in self.entrypoint_definition.inputs:
self.entrypoint_definition.validate_input(key=key, value=value)
elif not self.entrypoint_definition.legacy_params:
raise StepInterfaceError(
f"Unable to find parameter '{key}' in step function "
"signature."
)
if conflicting_parameters:
is_plural = "s" if len(conflicting_parameters) > 1 else ""
msg = f"Configured parameter{is_plural} for the step '{self.name}' conflict{'' if not is_plural else 's'} with parameter{is_plural} passed in runtime:\n"
for key, values in conflicting_parameters.items():
msg += (
f"`{key}`: config=`{values[0]}` | runtime=`{values[1]}`\n"
)
msg += """This happens, if you define values for step parameters in configuration file and pass same parameters from the code. Example:
```
# config.yaml
steps:
step_name:
parameters:
param_name: value1
# pipeline.py
@pipeline
def pipeline_():
step_name(param_name="other_value")
```
To avoid this consider setting step parameters only in one place (config or code).
"""
raise RuntimeError(msg)
def _validate_outputs(
self, outputs: Mapping[str, "PartialArtifactConfiguration"]
) -> None:
"""Validates the step output configuration.
Args:
outputs: The configured step outputs.
Raises:
StepInterfaceError: If an output for a non-existent name is
configured of an output artifact/materializer source does not
resolve to the correct class.
"""
allowed_output_names = set(self.entrypoint_definition.outputs)
for output_name, output in outputs.items():
if output_name not in allowed_output_names:
raise StepInterfaceError(
f"Got unexpected materializers for non-existent "
f"output '{output_name}' in step '{self.name}'. "
f"Only materializers for the outputs "
f"{allowed_output_names} of this step can"
f" be registered."
)
if output.materializer_source:
for source in output.materializer_source:
if not source_utils.validate_source_class(
source, expected_class=BaseMaterializer
):
raise StepInterfaceError(
f"Materializer source `{source}` "
f"for output '{output_name}' of step '{self.name}' "
"does not resolve to a `BaseMaterializer` subclass."
)
def _validate_inputs(
self,
input_artifacts: Dict[str, "StepArtifact"],
external_artifacts: Dict[str, "ExternalArtifactConfiguration"],
model_artifacts_or_metadata: Dict[str, "ModelVersionDataLazyLoader"],
client_lazy_loaders: Dict[str, "ClientLazyLoader"],
) -> None:
"""Validates the step inputs.
This method makes sure that all inputs are provided either as an
artifact or parameter.
Args:
input_artifacts: The input artifacts.
external_artifacts: The external input artifacts.
model_artifacts_or_metadata: The model artifacts or metadata.
client_lazy_loaders: The client lazy loaders.
Raises:
StepInterfaceError: If an entrypoint input is missing.
"""
for key in self.entrypoint_definition.inputs.keys():
if (
key in input_artifacts
or key in self.configuration.parameters
or key in external_artifacts
or key in model_artifacts_or_metadata
or key in client_lazy_loaders
):
continue
raise StepInterfaceError(
f"Missing entrypoint input '{key}' in step '{self.name}'."
)
def _finalize_configuration(
self,
input_artifacts: Dict[str, "StepArtifact"],
external_artifacts: Dict[str, "ExternalArtifactConfiguration"],
model_artifacts_or_metadata: Dict[str, "ModelVersionDataLazyLoader"],
client_lazy_loaders: Dict[str, "ClientLazyLoader"],
) -> "StepConfiguration":
"""Finalizes the configuration after the step was called.
Once the step was called, we know the outputs of previous steps
and that no additional user configurations will be made. That means
we can now collect the remaining artifact and materializer types
as well as check for the completeness of the step function parameters.
Args:
input_artifacts: The input artifacts of this step.
external_artifacts: The external artifacts of this step.
model_artifacts_or_metadata: The model artifacts or metadata of
this step.
client_lazy_loaders: The client lazy loaders of this step.
Raises:
StepInterfaceError: If explicit materializers were specified for an
output but they do not work for the data type(s) defined by
the type annotation.
Returns:
The finalized step configuration.
"""
from zenml.config.step_configurations import (
PartialArtifactConfiguration,
StepConfiguration,
StepConfigurationUpdate,
)
outputs: Dict[str, Dict[str, Union[Source, Tuple[Source, ...]]]] = (
defaultdict(dict)
)
for (
output_name,
output_annotation,
) in self.entrypoint_definition.outputs.items():
output = self._configuration.outputs.get(
output_name, PartialArtifactConfiguration()
)
if output.materializer_source:
# The materializer source was configured by the user. We
# validate that their configured materializer supports the
# output type. If the output annotation is a Union, we check
# that at least one of the specified materializers works with at
# least one of the types in the Union. If that's not the case,
# it would be a guaranteed failure at runtime and we fail early
# here.
if output_annotation.resolved_annotation is Any:
continue
materializer_classes: List[Type["BaseMaterializer"]] = [
source_utils.load(materializer_source)
for materializer_source in output.materializer_source
]
for data_type in output_annotation.get_output_types():
try:
materializer_utils.select_materializer(
data_type=data_type,
materializer_classes=materializer_classes,
)
break
except RuntimeError:
pass
else:
materializer_strings = [
materializer_source.import_path
for materializer_source in output.materializer_source
]
raise StepInterfaceError(
"Invalid materializers specified for output "
f"{output_name} of step {self.name}. None of the "
f"materializers ({materializer_strings}) are "
"able to save or load data of the type that is defined "
"for the output "
f"({output_annotation.resolved_annotation})."
)
else:
if output_annotation.resolved_annotation is Any:
outputs[output_name]["materializer_source"] = ()
outputs[output_name]["default_materializer_source"] = (
source_utils.resolve(
materializer_registry.get_default_materializer()
)
)
continue
materializer_sources = []
for output_type in output_annotation.get_output_types():
materializer_class = materializer_registry[output_type]
materializer_sources.append(
source_utils.resolve(materializer_class)
)
outputs[output_name]["materializer_source"] = tuple(
materializer_sources
)
parameters = self._finalize_parameters()
self.configure(parameters=parameters, merge=False)
self._validate_inputs(
input_artifacts=input_artifacts,
external_artifacts=external_artifacts,
model_artifacts_or_metadata=model_artifacts_or_metadata,
client_lazy_loaders=client_lazy_loaders,
)
values = dict_utils.remove_none_values({"outputs": outputs or None})
config = StepConfigurationUpdate(**values)
self._apply_configuration(config)
self._configuration = self._configuration.model_copy(
update={
"caching_parameters": self.caching_parameters,
"external_input_artifacts": external_artifacts,
"model_artifacts_or_metadata": model_artifacts_or_metadata,
"client_lazy_loaders": client_lazy_loaders,
}
)
return StepConfiguration.model_validate(
self._configuration.model_dump()
)
def _finalize_parameters(self) -> Dict[str, Any]:
"""Finalizes the config parameters for running this step.
Returns:
All parameter values for running this step.
"""
params = {}
for key, value in self.configuration.parameters.items():
if key not in self.entrypoint_definition.inputs:
continue
annotation = self.entrypoint_definition.inputs[key].annotation
annotation = resolve_type_annotation(annotation)
if inspect.isclass(annotation) and issubclass(
annotation, BaseModel
):
# Make sure we have all necessary values to instantiate the
# pydantic model later
model = annotation(**value)
params[key] = model.model_dump()
else:
params[key] = value
if self.entrypoint_definition.legacy_params:
legacy_params = self._finalize_legacy_parameters()
params[self.entrypoint_definition.legacy_params.name] = (
legacy_params
)
return params
def _finalize_legacy_parameters(self) -> Dict[str, Any]:
"""Verifies and prepares the config parameters for running this step.
When the step requires config parameters, this method:
- checks if config parameters were set via a config object or file
- tries to set missing config parameters from default values of the
config class
Returns:
Values for the previously unconfigured function parameters.
Raises:
MissingStepParameterError: If no value could be found for one or
more config parameters.
StepInterfaceError: If the parameter class validation failed.
"""
if not self.entrypoint_definition.legacy_params:
return {}
logger.warning(
"The `BaseParameters` class to define step parameters is "
"deprecated. Check out our docs "
"https://docs.zenml.io/how-to/use-configuration-files/how-to-use-config "
"for information on how to parameterize your steps. As a quick "
"fix to get rid of this warning, make sure your parameter class "
"inherits from `pydantic.BaseModel` instead of the "
"`BaseParameters` class."
)
# parameters for the `BaseParameters` class specified in the "new" way
# by specifying a dict of parameters for the corresponding key
params_defined_in_new_way = (
self.configuration.parameters.get(
self.entrypoint_definition.legacy_params.name
)
or {}
)
values = {}
missing_keys = []
for (
name,
field,
) in self.entrypoint_definition.legacy_params.annotation.model_fields.items():
if name in self.configuration.parameters:
# a value for this parameter has been set already
values[name] = self.configuration.parameters[name]
elif name in params_defined_in_new_way:
# a value for this parameter has been set in the "new" way
# already
values[name] = params_defined_in_new_way[name]
elif field.is_required():
# this field has no default value set and therefore needs
# to be passed via an initialized config object
missing_keys.append(name)
else:
# use default value from the pydantic config class
values[name] = field.default
if missing_keys:
raise MissingStepParameterError(
self.name,
missing_keys,
self.entrypoint_definition.legacy_params.annotation,
)
if (
getattr(
self.entrypoint_definition.legacy_params.annotation.model_config,
"extra",
None,
)
== "allow"
):
# Add all parameters for the config class for backwards
# compatibility if the config class allows extra attributes
values.update(self.configuration.parameters)
try:
self.entrypoint_definition.legacy_params.annotation(**values)
except ValidationError:
raise StepInterfaceError("Failed to validate function parameters.")
return values
caching_parameters: Dict[str, Any]
property
readonly
Caching parameters for this step.
Returns:
Type | Description |
---|---|
Dict[str, Any] |
A dictionary containing the caching parameters |
configuration: PartialStepConfiguration
property
readonly
The configuration of the step.
Returns:
Type | Description |
---|---|
PartialStepConfiguration |
The configuration of the step. |
docstring: Optional[str]
property
readonly
The docstring of this step.
Returns:
Type | Description |
---|---|
Optional[str] |
The docstring of this step. |
enable_cache: Optional[bool]
property
readonly
If caching is enabled for the step.
Returns:
Type | Description |
---|---|
Optional[bool] |
If caching is enabled for the step. |
name: str
property
readonly
The name of the step.
Returns:
Type | Description |
---|---|
str |
The name of the step. |
source_code: str
property
readonly
The source code of this step.
Returns:
Type | Description |
---|---|
str |
The source code of this step. |
source_object: Any
property
readonly
The source object of this step.
Returns:
Type | Description |
---|---|
Any |
The source object of this step. |
upstream_steps: Set[BaseStep]
property
readonly
Names of the upstream steps of this step.
This property will only contain the full set of upstream steps once
it's parent pipeline connect(...)
method was called.
Returns:
Type | Description |
---|---|
Set[BaseStep] |
Set of upstream step names. |
__call__(self, *args, *, id=None, after=None, **kwargs)
special
Handle a call of the step.
This method does one of two things: * If there is an active pipeline context, it adds an invocation of the step instance to the pipeline. * If no pipeline is active, it calls the step entrypoint function.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
*args |
Any |
Entrypoint function arguments. |
() |
id |
Optional[str] |
Invocation ID to use. |
None |
after |
Union[str, Sequence[str]] |
Upstream steps for the invocation. |
None |
**kwargs |
Any |
Entrypoint function keyword arguments. |
{} |
Returns:
Type | Description |
---|---|
Any |
The outputs of the entrypoint function call. |
Source code in zenml/steps/base_step.py
def __call__(
self,
*args: Any,
id: Optional[str] = None,
after: Union[str, Sequence[str], None] = None,
**kwargs: Any,
) -> Any:
"""Handle a call of the step.
This method does one of two things:
* If there is an active pipeline context, it adds an invocation of the
step instance to the pipeline.
* If no pipeline is active, it calls the step entrypoint function.
Args:
*args: Entrypoint function arguments.
id: Invocation ID to use.
after: Upstream steps for the invocation.
**kwargs: Entrypoint function keyword arguments.
Returns:
The outputs of the entrypoint function call.
"""
from zenml.new.pipelines.pipeline import Pipeline
if not Pipeline.ACTIVE_PIPELINE:
# The step is being called outside the context of a pipeline, either
# run the step function or run it as a single step pipeline on the
# active stack
run_without_stack = handle_bool_env_var(
ENV_ZENML_RUN_SINGLE_STEPS_WITHOUT_STACK, default=False
)
if run_without_stack:
return self.call_entrypoint(*args, **kwargs)
else:
return run_as_single_step_pipeline(self, *args, **kwargs)
(
input_artifacts,
external_artifacts,
model_artifacts_or_metadata,
client_lazy_loaders,
parameters,
default_parameters,
) = self._parse_call_args(*args, **kwargs)
upstream_steps = {
artifact.invocation_id for artifact in input_artifacts.values()
}
if isinstance(after, str):
upstream_steps.add(after)
elif isinstance(after, Sequence):
upstream_steps = upstream_steps.union(after)
invocation_id = Pipeline.ACTIVE_PIPELINE.add_step_invocation(
step=self,
input_artifacts=input_artifacts,
external_artifacts=external_artifacts,
model_artifacts_or_metadata=model_artifacts_or_metadata,
client_lazy_loaders=client_lazy_loaders,
parameters=parameters,
default_parameters=default_parameters,
upstream_steps=upstream_steps,
custom_id=id,
allow_id_suffix=not id,
)
outputs = []
for key, annotation in self.entrypoint_definition.outputs.items():
output = StepArtifact(
invocation_id=invocation_id,
output_name=key,
annotation=annotation,
pipeline=Pipeline.ACTIVE_PIPELINE,
)
outputs.append(output)
return outputs[0] if len(outputs) == 1 else outputs
__init__(self, *args, *, name=None, enable_cache=None, enable_artifact_metadata=None, enable_artifact_visualization=None, enable_step_logs=None, experiment_tracker=None, step_operator=None, parameters=None, output_materializers=None, settings=None, extra=None, on_failure=None, on_success=None, model=None, retry=None, **kwargs)
special
Initializes a step.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
*args |
Any |
Positional arguments passed to the step. |
() |
name |
Optional[str] |
The name of the step. |
None |
enable_cache |
Optional[bool] |
If caching should be enabled for this step. |
None |
enable_artifact_metadata |
Optional[bool] |
If artifact metadata should be enabled for this step. |
None |
enable_artifact_visualization |
Optional[bool] |
If artifact visualization should be enabled for this step. |
None |
enable_step_logs |
Optional[bool] |
Enable step logs for this step. |
None |
experiment_tracker |
Optional[str] |
The experiment tracker to use for this step. |
None |
step_operator |
Optional[str] |
The step operator to use for this step. |
None |
parameters |
Optional[ParametersOrDict] |
Function parameters for this step |
None |
output_materializers |
Optional[OutputMaterializersSpecification] |
Output materializers for this step. If given as a dict, the keys must be a subset of the output names of this step. If a single value (type or string) is given, the materializer will be used for all outputs. |
None |
settings |
Optional[Mapping[str, SettingsOrDict]] |
settings for this step. |
None |
extra |
Optional[Dict[str, Any]] |
Extra configurations for this step. |
None |
on_failure |
Optional[HookSpecification] |
Callback function in event of failure of the step. Can
be a function with a single argument of type |
None |
on_success |
Optional[HookSpecification] |
Callback function in event of success of the step. Can
be a function with no arguments, or a source path to such a
function (e.g. |
None |
model |
Optional[Model] |
configuration of the model version in the Model Control Plane. |
None |
retry |
Optional[zenml.config.retry_config.StepRetryConfig] |
Configuration for retrying the step in case of failure. |
None |
**kwargs |
Any |
Keyword arguments passed to the step. |
{} |
Source code in zenml/steps/base_step.py
def __init__(
self,
*args: Any,
name: Optional[str] = None,
enable_cache: Optional[bool] = None,
enable_artifact_metadata: Optional[bool] = None,
enable_artifact_visualization: Optional[bool] = None,
enable_step_logs: Optional[bool] = None,
experiment_tracker: Optional[str] = None,
step_operator: Optional[str] = None,
parameters: Optional["ParametersOrDict"] = None,
output_materializers: Optional[
"OutputMaterializersSpecification"
] = None,
settings: Optional[Mapping[str, "SettingsOrDict"]] = None,
extra: Optional[Dict[str, Any]] = None,
on_failure: Optional["HookSpecification"] = None,
on_success: Optional["HookSpecification"] = None,
model: Optional["Model"] = None,
retry: Optional[StepRetryConfig] = None,
**kwargs: Any,
) -> None:
"""Initializes a step.
Args:
*args: Positional arguments passed to the step.
name: The name of the step.
enable_cache: If caching should be enabled for this step.
enable_artifact_metadata: If artifact metadata should be enabled
for this step.
enable_artifact_visualization: If artifact visualization should be
enabled for this step.
enable_step_logs: Enable step logs for this step.
experiment_tracker: The experiment tracker to use for this step.
step_operator: The step operator to use for this step.
parameters: Function parameters for this step
output_materializers: Output materializers for this step. If
given as a dict, the keys must be a subset of the output names
of this step. If a single value (type or string) is given, the
materializer will be used for all outputs.
settings: settings for this step.
extra: Extra configurations for this step.
on_failure: Callback function in event of failure of the step. Can
be a function with a single argument of type `BaseException`, or
a source path to such a function (e.g. `module.my_function`).
on_success: Callback function in event of success of the step. Can
be a function with no arguments, or a source path to such a
function (e.g. `module.my_function`).
model: configuration of the model version in the Model Control Plane.
retry: Configuration for retrying the step in case of failure.
**kwargs: Keyword arguments passed to the step.
"""
from zenml.config.step_configurations import PartialStepConfiguration
self._upstream_steps: Set["BaseStep"] = set()
self.entrypoint_definition = validate_entrypoint_function(
self.entrypoint, reserved_arguments=["after", "id"]
)
name = name or self.__class__.__name__
requires_context = self.entrypoint_definition.context is not None
if enable_cache is None:
if requires_context:
# Using the StepContext inside a step provides access to
# external resources which might influence the step execution.
# We therefore disable caching unless it is explicitly enabled
enable_cache = False
logger.debug(
"Step `%s`: Step context required and caching not "
"explicitly enabled.",
name,
)
logger.debug(
"Step `%s`: Caching %s.",
name,
"enabled" if enable_cache is not False else "disabled",
)
logger.debug(
"Step `%s`: Artifact metadata %s.",
name,
"enabled" if enable_artifact_metadata is not False else "disabled",
)
logger.debug(
"Step `%s`: Artifact visualization %s.",
name,
"enabled"
if enable_artifact_visualization is not False
else "disabled",
)
logger.debug(
"Step `%s`: logs %s.",
name,
"enabled" if enable_step_logs is not False else "disabled",
)
if model is not None:
logger.debug(
"Step `%s`: Is in Model context %s.",
name,
{
"model": model.name,
"version": model.version,
},
)
self._configuration = PartialStepConfiguration(
name=name,
enable_cache=enable_cache,
enable_artifact_metadata=enable_artifact_metadata,
enable_artifact_visualization=enable_artifact_visualization,
enable_step_logs=enable_step_logs,
)
self.configure(
experiment_tracker=experiment_tracker,
step_operator=step_operator,
output_materializers=output_materializers,
parameters=parameters,
settings=settings,
extra=extra,
on_failure=on_failure,
on_success=on_success,
model=model,
retry=retry,
)
self._verify_and_apply_init_params(*args, **kwargs)
notebook_utils.try_to_save_notebook_cell_code(self.source_object)
after(self, step)
Adds an upstream step to this step.
Calling this method makes sure this step only starts running once the given step has successfully finished executing.
Note: This can only be called inside the pipeline connect function
which is decorated with the @pipeline
decorator. Any calls outside
this function will be ignored.
Examples:
The following pipeline will run its steps sequentially in the following order: step_2 -> step_1 -> step_3
@pipeline
def example_pipeline(step_1, step_2, step_3):
step_1.after(step_2)
step_3(step_1(), step_2())
Parameters:
Name | Type | Description | Default |
---|---|---|---|
step |
BaseStep |
A step which should finish executing before this step is started. |
required |
Source code in zenml/steps/base_step.py
def after(self, step: "BaseStep") -> None:
"""Adds an upstream step to this step.
Calling this method makes sure this step only starts running once the
given step has successfully finished executing.
**Note**: This can only be called inside the pipeline connect function
which is decorated with the `@pipeline` decorator. Any calls outside
this function will be ignored.
Example:
The following pipeline will run its steps sequentially in the following
order: step_2 -> step_1 -> step_3
```python
@pipeline
def example_pipeline(step_1, step_2, step_3):
step_1.after(step_2)
step_3(step_1(), step_2())
```
Args:
step: A step which should finish executing before this step is
started.
"""
self._upstream_steps.add(step)
call_entrypoint(self, *args, **kwargs)
Calls the entrypoint function of the step.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
*args |
Any |
Entrypoint function arguments. |
() |
**kwargs |
Any |
Entrypoint function keyword arguments. |
{} |
Returns:
Type | Description |
---|---|
Any |
The return value of the entrypoint function. |
Exceptions:
Type | Description |
---|---|
StepInterfaceError |
If the arguments to the entrypoint function are invalid. |
Source code in zenml/steps/base_step.py
def call_entrypoint(self, *args: Any, **kwargs: Any) -> Any:
"""Calls the entrypoint function of the step.
Args:
*args: Entrypoint function arguments.
**kwargs: Entrypoint function keyword arguments.
Returns:
The return value of the entrypoint function.
Raises:
StepInterfaceError: If the arguments to the entrypoint function are
invalid.
"""
try:
validated_args = pydantic_utils.validate_function_args(
self.entrypoint,
ConfigDict(arbitrary_types_allowed=True),
*args,
**kwargs,
)
except ValidationError as e:
raise StepInterfaceError(
"Invalid step function entrypoint arguments. Check out the "
"pydantic error above for more details."
) from e
return self.entrypoint(**validated_args)
configure(self, name=None, enable_cache=None, enable_artifact_metadata=None, enable_artifact_visualization=None, enable_step_logs=None, experiment_tracker=None, step_operator=None, parameters=None, output_materializers=None, settings=None, extra=None, on_failure=None, on_success=None, model=None, merge=True, retry=None)
Configures the step.
Configuration merging example:
* merge==True
:
step.configure(extra={"key1": 1})
step.configure(extra={"key2": 2}, merge=True)
step.configuration.extra # {"key1": 1, "key2": 2}
* merge==False
:
step.configure(extra={"key1": 1})
step.configure(extra={"key2": 2}, merge=False)
step.configuration.extra # {"key2": 2}
Parameters:
Name | Type | Description | Default |
---|---|---|---|
name |
Optional[str] |
DEPRECATED: The name of the step. |
None |
enable_cache |
Optional[bool] |
If caching should be enabled for this step. |
None |
enable_artifact_metadata |
Optional[bool] |
If artifact metadata should be enabled for this step. |
None |
enable_artifact_visualization |
Optional[bool] |
If artifact visualization should be enabled for this step. |
None |
enable_step_logs |
Optional[bool] |
If step logs should be enabled for this step. |
None |
experiment_tracker |
Optional[str] |
The experiment tracker to use for this step. |
None |
step_operator |
Optional[str] |
The step operator to use for this step. |
None |
parameters |
Optional[ParametersOrDict] |
Function parameters for this step |
None |
output_materializers |
Optional[OutputMaterializersSpecification] |
Output materializers for this step. If given as a dict, the keys must be a subset of the output names of this step. If a single value (type or string) is given, the materializer will be used for all outputs. |
None |
settings |
Optional[Mapping[str, SettingsOrDict]] |
settings for this step. |
None |
extra |
Optional[Dict[str, Any]] |
Extra configurations for this step. |
None |
on_failure |
Optional[HookSpecification] |
Callback function in event of failure of the step. Can
be a function with a single argument of type |
None |
on_success |
Optional[HookSpecification] |
Callback function in event of success of the step. Can
be a function with no arguments, or a source path to such a
function (e.g. |
None |
model |
Optional[Model] |
configuration of the model version in the Model Control Plane. |
None |
merge |
bool |
If |
True |
retry |
Optional[zenml.config.retry_config.StepRetryConfig] |
Configuration for retrying the step in case of failure. |
None |
Returns:
Type | Description |
---|---|
~T |
The step instance that this method was called on. |
Source code in zenml/steps/base_step.py
def configure(
self: T,
name: Optional[str] = None,
enable_cache: Optional[bool] = None,
enable_artifact_metadata: Optional[bool] = None,
enable_artifact_visualization: Optional[bool] = None,
enable_step_logs: Optional[bool] = None,
experiment_tracker: Optional[str] = None,
step_operator: Optional[str] = None,
parameters: Optional["ParametersOrDict"] = None,
output_materializers: Optional[
"OutputMaterializersSpecification"
] = None,
settings: Optional[Mapping[str, "SettingsOrDict"]] = None,
extra: Optional[Dict[str, Any]] = None,
on_failure: Optional["HookSpecification"] = None,
on_success: Optional["HookSpecification"] = None,
model: Optional["Model"] = None,
merge: bool = True,
retry: Optional[StepRetryConfig] = None,
) -> T:
"""Configures the step.
Configuration merging example:
* `merge==True`:
step.configure(extra={"key1": 1})
step.configure(extra={"key2": 2}, merge=True)
step.configuration.extra # {"key1": 1, "key2": 2}
* `merge==False`:
step.configure(extra={"key1": 1})
step.configure(extra={"key2": 2}, merge=False)
step.configuration.extra # {"key2": 2}
Args:
name: DEPRECATED: The name of the step.
enable_cache: If caching should be enabled for this step.
enable_artifact_metadata: If artifact metadata should be enabled
for this step.
enable_artifact_visualization: If artifact visualization should be
enabled for this step.
enable_step_logs: If step logs should be enabled for this step.
experiment_tracker: The experiment tracker to use for this step.
step_operator: The step operator to use for this step.
parameters: Function parameters for this step
output_materializers: Output materializers for this step. If
given as a dict, the keys must be a subset of the output names
of this step. If a single value (type or string) is given, the
materializer will be used for all outputs.
settings: settings for this step.
extra: Extra configurations for this step.
on_failure: Callback function in event of failure of the step. Can
be a function with a single argument of type `BaseException`, or
a source path to such a function (e.g. `module.my_function`).
on_success: Callback function in event of success of the step. Can
be a function with no arguments, or a source path to such a
function (e.g. `module.my_function`).
model: configuration of the model version in the Model Control Plane.
merge: If `True`, will merge the given dictionary configurations
like `parameters` and `settings` with existing
configurations. If `False` the given configurations will
overwrite all existing ones. See the general description of this
method for an example.
retry: Configuration for retrying the step in case of failure.
Returns:
The step instance that this method was called on.
"""
from zenml.config.step_configurations import StepConfigurationUpdate
from zenml.hooks.hook_validators import resolve_and_validate_hook
if name:
logger.warning("Configuring the name of a step is deprecated.")
def _resolve_if_necessary(
value: Union[str, Source, Type[Any]],
) -> Source:
if isinstance(value, str):
return Source.from_import_path(value)
elif isinstance(value, Source):
return value
else:
return source_utils.resolve(value)
def _convert_to_tuple(value: Any) -> Tuple[Source, ...]:
if isinstance(value, str) or not isinstance(value, Sequence):
return (_resolve_if_necessary(value),)
else:
return tuple(_resolve_if_necessary(v) for v in value)
outputs: Dict[str, Dict[str, Tuple[Source, ...]]] = defaultdict(dict)
allowed_output_names = set(self.entrypoint_definition.outputs)
if output_materializers:
if not isinstance(output_materializers, Mapping):
sources = _convert_to_tuple(output_materializers)
output_materializers = {
output_name: sources
for output_name in allowed_output_names
}
for output_name, materializer in output_materializers.items():
sources = _convert_to_tuple(materializer)
outputs[output_name]["materializer_source"] = sources
failure_hook_source = None
if on_failure:
# string of on_failure hook function to be used for this step
failure_hook_source = resolve_and_validate_hook(on_failure)
success_hook_source = None
if on_success:
# string of on_success hook function to be used for this step
success_hook_source = resolve_and_validate_hook(on_success)
if isinstance(parameters, BaseParameters):
parameters = parameters.model_dump()
values = dict_utils.remove_none_values(
{
"enable_cache": enable_cache,
"enable_artifact_metadata": enable_artifact_metadata,
"enable_artifact_visualization": enable_artifact_visualization,
"enable_step_logs": enable_step_logs,
"experiment_tracker": experiment_tracker,
"step_operator": step_operator,
"parameters": parameters,
"settings": settings,
"outputs": outputs or None,
"extra": extra,
"failure_hook_source": failure_hook_source,
"success_hook_source": success_hook_source,
"model": model,
"retry": retry,
}
)
config = StepConfigurationUpdate(**values)
self._apply_configuration(config, merge=merge)
return self
copy(self)
Copies the step.
Returns:
Type | Description |
---|---|
BaseStep |
The step copy. |
Source code in zenml/steps/base_step.py
def copy(self) -> "BaseStep":
"""Copies the step.
Returns:
The step copy.
"""
return copy.deepcopy(self)
entrypoint(self, *args, **kwargs)
Abstract method for core step logic.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
*args |
Any |
Positional arguments passed to the step. |
() |
**kwargs |
Any |
Keyword arguments passed to the step. |
{} |
Returns:
Type | Description |
---|---|
Any |
The output of the step. |
Source code in zenml/steps/base_step.py
@abstractmethod
def entrypoint(self, *args: Any, **kwargs: Any) -> Any:
"""Abstract method for core step logic.
Args:
*args: Positional arguments passed to the step.
**kwargs: Keyword arguments passed to the step.
Returns:
The output of the step.
"""
load_from_source(source)
classmethod
Loads a step from source.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
source |
Union[zenml.config.source.Source, str] |
The path to the step source. |
required |
Returns:
Type | Description |
---|---|
BaseStep |
The loaded step. |
Exceptions:
Type | Description |
---|---|
ValueError |
If the source is not a valid step source. |
Source code in zenml/steps/base_step.py
@classmethod
def load_from_source(cls, source: Union[Source, str]) -> "BaseStep":
"""Loads a step from source.
Args:
source: The path to the step source.
Returns:
The loaded step.
Raises:
ValueError: If the source is not a valid step source.
"""
obj = source_utils.load(source)
if isinstance(obj, BaseStep):
return obj
elif isinstance(obj, type) and issubclass(obj, BaseStep):
return obj()
else:
raise ValueError("Invalid step source.")
resolve(self)
Resolves the step.
Returns:
Type | Description |
---|---|
Source |
The step source. |
Source code in zenml/steps/base_step.py
def resolve(self) -> Source:
"""Resolves the step.
Returns:
The step source.
"""
return source_utils.resolve(self.__class__)
with_options(self, enable_cache=None, enable_artifact_metadata=None, enable_artifact_visualization=None, enable_step_logs=None, experiment_tracker=None, step_operator=None, parameters=None, output_materializers=None, settings=None, extra=None, on_failure=None, on_success=None, model=None, merge=True)
Copies the step and applies the given configurations.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
enable_cache |
Optional[bool] |
If caching should be enabled for this step. |
None |
enable_artifact_metadata |
Optional[bool] |
If artifact metadata should be enabled for this step. |
None |
enable_artifact_visualization |
Optional[bool] |
If artifact visualization should be enabled for this step. |
None |
enable_step_logs |
Optional[bool] |
If step logs should be enabled for this step. |
None |
experiment_tracker |
Optional[str] |
The experiment tracker to use for this step. |
None |
step_operator |
Optional[str] |
The step operator to use for this step. |
None |
parameters |
Optional[ParametersOrDict] |
Function parameters for this step |
None |
output_materializers |
Optional[OutputMaterializersSpecification] |
Output materializers for this step. If given as a dict, the keys must be a subset of the output names of this step. If a single value (type or string) is given, the materializer will be used for all outputs. |
None |
settings |
Optional[Mapping[str, SettingsOrDict]] |
settings for this step. |
None |
extra |
Optional[Dict[str, Any]] |
Extra configurations for this step. |
None |
on_failure |
Optional[HookSpecification] |
Callback function in event of failure of the step. Can
be a function with a single argument of type |
None |
on_success |
Optional[HookSpecification] |
Callback function in event of success of the step. Can
be a function with no arguments, or a source path to such a
function (e.g. |
None |
model |
Optional[Model] |
configuration of the model version in the Model Control Plane. |
None |
merge |
bool |
If |
True |
Returns:
Type | Description |
---|---|
BaseStep |
The copied step instance. |
Source code in zenml/steps/base_step.py
def with_options(
self,
enable_cache: Optional[bool] = None,
enable_artifact_metadata: Optional[bool] = None,
enable_artifact_visualization: Optional[bool] = None,
enable_step_logs: Optional[bool] = None,
experiment_tracker: Optional[str] = None,
step_operator: Optional[str] = None,
parameters: Optional["ParametersOrDict"] = None,
output_materializers: Optional[
"OutputMaterializersSpecification"
] = None,
settings: Optional[Mapping[str, "SettingsOrDict"]] = None,
extra: Optional[Dict[str, Any]] = None,
on_failure: Optional["HookSpecification"] = None,
on_success: Optional["HookSpecification"] = None,
model: Optional["Model"] = None,
merge: bool = True,
) -> "BaseStep":
"""Copies the step and applies the given configurations.
Args:
enable_cache: If caching should be enabled for this step.
enable_artifact_metadata: If artifact metadata should be enabled
for this step.
enable_artifact_visualization: If artifact visualization should be
enabled for this step.
enable_step_logs: If step logs should be enabled for this step.
experiment_tracker: The experiment tracker to use for this step.
step_operator: The step operator to use for this step.
parameters: Function parameters for this step
output_materializers: Output materializers for this step. If
given as a dict, the keys must be a subset of the output names
of this step. If a single value (type or string) is given, the
materializer will be used for all outputs.
settings: settings for this step.
extra: Extra configurations for this step.
on_failure: Callback function in event of failure of the step. Can
be a function with a single argument of type `BaseException`, or
a source path to such a function (e.g. `module.my_function`).
on_success: Callback function in event of success of the step. Can
be a function with no arguments, or a source path to such a
function (e.g. `module.my_function`).
model: configuration of the model version in the Model Control Plane.
merge: If `True`, will merge the given dictionary configurations
like `parameters` and `settings` with existing
configurations. If `False` the given configurations will
overwrite all existing ones. See the general description of this
method for an example.
Returns:
The copied step instance.
"""
step_copy = self.copy()
step_copy.configure(
enable_cache=enable_cache,
enable_artifact_metadata=enable_artifact_metadata,
enable_artifact_visualization=enable_artifact_visualization,
enable_step_logs=enable_step_logs,
experiment_tracker=experiment_tracker,
step_operator=step_operator,
parameters=parameters,
output_materializers=output_materializers,
settings=settings,
extra=extra,
on_failure=on_failure,
on_success=on_success,
model=model,
merge=merge,
)
return step_copy
BaseStepMeta (type)
Metaclass for BaseStep
.
Makes sure that the entrypoint function has valid parameters and type annotations.
Source code in zenml/steps/base_step.py
class BaseStepMeta(type):
"""Metaclass for `BaseStep`.
Makes sure that the entrypoint function has valid parameters and type
annotations.
"""
def __new__(
mcs, name: str, bases: Tuple[Type[Any], ...], dct: Dict[str, Any]
) -> "BaseStepMeta":
"""Set up a new class with a qualified spec.
Args:
name: The name of the class.
bases: The base classes of the class.
dct: The attributes of the class.
Returns:
The new class.
"""
cls = cast(Type["BaseStep"], super().__new__(mcs, name, bases, dct))
if name not in {"BaseStep", "_DecoratedStep"}:
validate_entrypoint_function(cls.entrypoint)
return cls
__new__(mcs, name, bases, dct)
special
staticmethod
Set up a new class with a qualified spec.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
name |
str |
The name of the class. |
required |
bases |
Tuple[Type[Any], ...] |
The base classes of the class. |
required |
dct |
Dict[str, Any] |
The attributes of the class. |
required |
Returns:
Type | Description |
---|---|
BaseStepMeta |
The new class. |
Source code in zenml/steps/base_step.py
def __new__(
mcs, name: str, bases: Tuple[Type[Any], ...], dct: Dict[str, Any]
) -> "BaseStepMeta":
"""Set up a new class with a qualified spec.
Args:
name: The name of the class.
bases: The base classes of the class.
dct: The attributes of the class.
Returns:
The new class.
"""
cls = cast(Type["BaseStep"], super().__new__(mcs, name, bases, dct))
if name not in {"BaseStep", "_DecoratedStep"}:
validate_entrypoint_function(cls.entrypoint)
return cls
entrypoint_function_utils
Util functions for step and pipeline entrypoint functions.
EntrypointFunctionDefinition (tuple)
Class representing a step entrypoint function.
Attributes:
Name | Type | Description |
---|---|---|
inputs |
Dict[str, inspect.Parameter] |
The entrypoint function inputs. |
outputs |
Dict[str, zenml.steps.utils.OutputSignature] |
The entrypoint function outputs. This dictionary maps output names to output annotations. |
context |
Optional[inspect.Parameter] |
Optional parameter representing the |
legacy_params |
Optional[inspect.Parameter] |
Optional parameter representing the |
Source code in zenml/steps/entrypoint_function_utils.py
class EntrypointFunctionDefinition(NamedTuple):
"""Class representing a step entrypoint function.
Attributes:
inputs: The entrypoint function inputs.
outputs: The entrypoint function outputs. This dictionary maps output
names to output annotations.
context: Optional parameter representing the `StepContext` input.
legacy_params: Optional parameter representing the `BaseParameters`
input.
"""
inputs: Dict[str, inspect.Parameter]
outputs: Dict[str, OutputSignature]
context: Optional[inspect.Parameter]
legacy_params: Optional[inspect.Parameter]
def validate_input(self, key: str, value: Any) -> None:
"""Validates an input to the step entrypoint function.
Args:
key: The key for which the input was passed
value: The input value.
Raises:
KeyError: If the function has no input for the given key.
RuntimeError: If a parameter is passed for an input that is
annotated as an `UnmaterializedArtifact`.
RuntimeError: If the input value is not valid for the type
annotation provided for the function parameter.
StepInterfaceError: If the input is a parameter and not JSON
serializable.
"""
from zenml.artifacts.external_artifact import ExternalArtifact
from zenml.artifacts.unmaterialized_artifact import (
UnmaterializedArtifact,
)
from zenml.client_lazy_loader import ClientLazyLoader
from zenml.models import (
ArtifactVersionResponse,
RunMetadataResponse,
)
if key not in self.inputs:
raise KeyError(
f"Received step entrypoint input for invalid key {key}."
)
parameter = self.inputs[key]
if isinstance(
value,
(
StepArtifact,
ExternalArtifact,
ArtifactVersionResponse,
RunMetadataResponse,
ClientLazyLoader,
),
):
# If we were to do any type validation for artifacts here, we
# would not be able to leverage pydantics type coercion (e.g.
# providing an `int` artifact for a `float` input)
return
# Not an artifact -> This is a parameter
if parameter.annotation is UnmaterializedArtifact:
raise RuntimeError(
"Passing parameter for input of type `UnmaterializedArtifact` "
"is not allowed."
)
if not yaml_utils.is_json_serializable(value):
raise StepInterfaceError(
f"Argument type (`{type(value)}`) for argument "
f"'{key}' is not JSON serializable and can not be passed as "
"a parameter. This input can either be provided by the "
"output of another step or as an external artifact: "
"https://docs.zenml.io/user-guide/starter-guide/manage-artifacts#managing-artifacts-not-produced-by-zenml-pipelines"
)
try:
self._validate_input_value(parameter=parameter, value=value)
except ValidationError as e:
raise RuntimeError(
f"Input validation failed for input '{parameter.name}': "
f"Expected type `{parameter.annotation}` but received type "
f"`{type(value)}`."
) from e
def _validate_input_value(
self, parameter: inspect.Parameter, value: Any
) -> None:
"""Validates an input value to the step entrypoint function.
Args:
parameter: The function parameter for which the value was provided.
value: The input value.
"""
config_dict = ConfigDict(arbitrary_types_allowed=False)
# Create a pydantic model with just a single required field with the
# type annotation of the parameter to verify the input type including
# pydantics type coercion
validation_model_class = create_model(
"input_validation_model",
__config__=config_dict,
value=(parameter.annotation, ...),
)
validation_model_class(value=value)
__getnewargs__(self)
special
Return self as a plain tuple. Used by copy and pickle.
Source code in zenml/steps/entrypoint_function_utils.py
def __getnewargs__(self):
'Return self as a plain tuple. Used by copy and pickle.'
return _tuple(self)
__new__(_cls, inputs, outputs, context, legacy_params)
special
staticmethod
Create new instance of EntrypointFunctionDefinition(inputs, outputs, context, legacy_params)
__repr__(self)
special
Return a nicely formatted representation string
Source code in zenml/steps/entrypoint_function_utils.py
def __repr__(self):
'Return a nicely formatted representation string'
return self.__class__.__name__ + repr_fmt % self
validate_input(self, key, value)
Validates an input to the step entrypoint function.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
key |
str |
The key for which the input was passed |
required |
value |
Any |
The input value. |
required |
Exceptions:
Type | Description |
---|---|
KeyError |
If the function has no input for the given key. |
RuntimeError |
If a parameter is passed for an input that is
annotated as an |
RuntimeError |
If the input value is not valid for the type annotation provided for the function parameter. |
StepInterfaceError |
If the input is a parameter and not JSON serializable. |
Source code in zenml/steps/entrypoint_function_utils.py
def validate_input(self, key: str, value: Any) -> None:
"""Validates an input to the step entrypoint function.
Args:
key: The key for which the input was passed
value: The input value.
Raises:
KeyError: If the function has no input for the given key.
RuntimeError: If a parameter is passed for an input that is
annotated as an `UnmaterializedArtifact`.
RuntimeError: If the input value is not valid for the type
annotation provided for the function parameter.
StepInterfaceError: If the input is a parameter and not JSON
serializable.
"""
from zenml.artifacts.external_artifact import ExternalArtifact
from zenml.artifacts.unmaterialized_artifact import (
UnmaterializedArtifact,
)
from zenml.client_lazy_loader import ClientLazyLoader
from zenml.models import (
ArtifactVersionResponse,
RunMetadataResponse,
)
if key not in self.inputs:
raise KeyError(
f"Received step entrypoint input for invalid key {key}."
)
parameter = self.inputs[key]
if isinstance(
value,
(
StepArtifact,
ExternalArtifact,
ArtifactVersionResponse,
RunMetadataResponse,
ClientLazyLoader,
),
):
# If we were to do any type validation for artifacts here, we
# would not be able to leverage pydantics type coercion (e.g.
# providing an `int` artifact for a `float` input)
return
# Not an artifact -> This is a parameter
if parameter.annotation is UnmaterializedArtifact:
raise RuntimeError(
"Passing parameter for input of type `UnmaterializedArtifact` "
"is not allowed."
)
if not yaml_utils.is_json_serializable(value):
raise StepInterfaceError(
f"Argument type (`{type(value)}`) for argument "
f"'{key}' is not JSON serializable and can not be passed as "
"a parameter. This input can either be provided by the "
"output of another step or as an external artifact: "
"https://docs.zenml.io/user-guide/starter-guide/manage-artifacts#managing-artifacts-not-produced-by-zenml-pipelines"
)
try:
self._validate_input_value(parameter=parameter, value=value)
except ValidationError as e:
raise RuntimeError(
f"Input validation failed for input '{parameter.name}': "
f"Expected type `{parameter.annotation}` but received type "
f"`{type(value)}`."
) from e
StepArtifact
Class to represent step output artifacts.
Source code in zenml/steps/entrypoint_function_utils.py
class StepArtifact:
"""Class to represent step output artifacts."""
def __init__(
self,
invocation_id: str,
output_name: str,
annotation: Any,
pipeline: "Pipeline",
) -> None:
"""Initialize a step artifact.
Args:
invocation_id: The ID of the invocation that produces this artifact.
output_name: The name of the output that produces this artifact.
annotation: The output type annotation.
pipeline: The pipeline which the invocation is part of.
"""
self.invocation_id = invocation_id
self.output_name = output_name
self.annotation = annotation
self.pipeline = pipeline
def __iter__(self) -> NoReturn:
"""Raise a custom error if someone is trying to iterate this object.
Raises:
StepInterfaceError: If trying to iterate this object.
"""
raise StepInterfaceError(
"Unable to unpack step artifact. This error is probably because "
"you're trying to unpack the return value of your step but the "
"step only returns a single artifact. For more information on how "
"to add type annotations to your step to indicate multiple "
"artifacts visit https://docs.zenml.io/how-to/build-pipelines/step-output-typing-and-annotation#type-annotations."
)
__init__(self, invocation_id, output_name, annotation, pipeline)
special
Initialize a step artifact.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
invocation_id |
str |
The ID of the invocation that produces this artifact. |
required |
output_name |
str |
The name of the output that produces this artifact. |
required |
annotation |
Any |
The output type annotation. |
required |
pipeline |
Pipeline |
The pipeline which the invocation is part of. |
required |
Source code in zenml/steps/entrypoint_function_utils.py
def __init__(
self,
invocation_id: str,
output_name: str,
annotation: Any,
pipeline: "Pipeline",
) -> None:
"""Initialize a step artifact.
Args:
invocation_id: The ID of the invocation that produces this artifact.
output_name: The name of the output that produces this artifact.
annotation: The output type annotation.
pipeline: The pipeline which the invocation is part of.
"""
self.invocation_id = invocation_id
self.output_name = output_name
self.annotation = annotation
self.pipeline = pipeline
__iter__(self)
special
Raise a custom error if someone is trying to iterate this object.
Exceptions:
Type | Description |
---|---|
StepInterfaceError |
If trying to iterate this object. |
Source code in zenml/steps/entrypoint_function_utils.py
def __iter__(self) -> NoReturn:
"""Raise a custom error if someone is trying to iterate this object.
Raises:
StepInterfaceError: If trying to iterate this object.
"""
raise StepInterfaceError(
"Unable to unpack step artifact. This error is probably because "
"you're trying to unpack the return value of your step but the "
"step only returns a single artifact. For more information on how "
"to add type annotations to your step to indicate multiple "
"artifacts visit https://docs.zenml.io/how-to/build-pipelines/step-output-typing-and-annotation#type-annotations."
)
get_step_entrypoint_signature(step)
Get the entrypoint signature of a step.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
step |
BaseStep |
The step for which to get the entrypoint signature. |
required |
Returns:
Type | Description |
---|---|
Signature |
The entrypoint function signature. |
Source code in zenml/steps/entrypoint_function_utils.py
def get_step_entrypoint_signature(step: "BaseStep") -> inspect.Signature:
"""Get the entrypoint signature of a step.
Args:
step: The step for which to get the entrypoint signature.
Returns:
The entrypoint function signature.
"""
from zenml.steps import BaseParameters, StepContext
signature = inspect.signature(step.entrypoint, follow_wrapped=True)
def _is_param_of_class(annotation: Any, class_: Type[Any]) -> bool:
annotation = resolve_type_annotation(annotation)
return inspect.isclass(annotation) and issubclass(annotation, class_)
parameters = list(signature.parameters.values())
# Filter out deprecated args: step context and legacy parameters
parameters = [
param
for param in parameters
if not _is_param_of_class(param.annotation, class_=BaseParameters)
and not _is_param_of_class(param.annotation, class_=StepContext)
]
signature = signature.replace(parameters=parameters)
return signature
validate_entrypoint_function(func, reserved_arguments=())
Validates a step entrypoint function.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
func |
Callable[..., Any] |
The step entrypoint function to validate. |
required |
reserved_arguments |
Sequence[str] |
The reserved arguments for the entrypoint function. |
() |
Exceptions:
Type | Description |
---|---|
StepInterfaceError |
If the entrypoint function has variable arguments or keyword arguments. |
StepInterfaceError |
If the entrypoint function has multiple
|
StepInterfaceError |
If the entrypoint function has multiple
|
RuntimeError |
If type annotations should be enforced and a type annotation is missing. |
Returns:
Type | Description |
---|---|
EntrypointFunctionDefinition |
A validated definition of the entrypoint function. |
Source code in zenml/steps/entrypoint_function_utils.py
def validate_entrypoint_function(
func: Callable[..., Any], reserved_arguments: Sequence[str] = ()
) -> EntrypointFunctionDefinition:
"""Validates a step entrypoint function.
Args:
func: The step entrypoint function to validate.
reserved_arguments: The reserved arguments for the entrypoint function.
Raises:
StepInterfaceError: If the entrypoint function has variable arguments
or keyword arguments.
StepInterfaceError: If the entrypoint function has multiple
`BaseParameter` arguments.
StepInterfaceError: If the entrypoint function has multiple
`StepContext` arguments.
RuntimeError: If type annotations should be enforced and a type
annotation is missing.
Returns:
A validated definition of the entrypoint function.
"""
from zenml.steps import BaseParameters, StepContext
signature = inspect.signature(func, follow_wrapped=True)
validate_reserved_arguments(
signature=signature, reserved_arguments=reserved_arguments
)
inputs = {}
context: Optional[inspect.Parameter] = None
legacy_params: Optional[inspect.Parameter] = None
signature_parameters = list(signature.parameters.items())
if signature_parameters and signature_parameters[0][0] == "self":
# TODO: Once we get rid of the old step decorator, we can also remove
# the `BaseStepMeta` class which right now calls this function on an
# unbound instance method when using the class-based API. If we get rid
# of that, this check and removal of the `self` parameter is not
# necessary anymore
signature_parameters = signature_parameters[1:]
for key, parameter in signature_parameters:
if parameter.kind in {parameter.VAR_POSITIONAL, parameter.VAR_KEYWORD}:
raise StepInterfaceError(
f"Variable args or kwargs not allowed for function "
f"{func.__name__}."
)
annotation = parameter.annotation
if annotation is parameter.empty:
if ENFORCE_TYPE_ANNOTATIONS:
raise RuntimeError(
f"Missing type annotation for input '{key}' of step "
f"function '{func.__name__}'."
)
# If a type annotation is missing, use `Any` instead
parameter = parameter.replace(annotation=Any)
annotation = resolve_type_annotation(annotation)
if inspect.isclass(annotation) and issubclass(
annotation, BaseParameters
):
if legacy_params is not None:
raise StepInterfaceError(
f"Found multiple parameter arguments "
f"('{legacy_params.name}' and '{key}') "
f"for function {func.__name__}."
)
legacy_params = parameter
elif inspect.isclass(annotation) and issubclass(
annotation, StepContext
):
if context is not None:
raise StepInterfaceError(
f"Found multiple context arguments "
f"('{context.name}' and '{key}') "
f"for function {func.__name__}."
)
context = parameter
else:
inputs[key] = parameter
outputs = parse_return_type_annotations(
func=func, enforce_type_annotations=ENFORCE_TYPE_ANNOTATIONS
)
return EntrypointFunctionDefinition(
inputs=inputs,
outputs=outputs,
context=context,
legacy_params=legacy_params,
)
validate_reserved_arguments(signature, reserved_arguments)
Validates that the signature does not contain any reserved arguments.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
signature |
Signature |
The signature to validate. |
required |
reserved_arguments |
Sequence[str] |
The reserved arguments for the signature. |
required |
Exceptions:
Type | Description |
---|---|
RuntimeError |
If the signature contains a reserved argument. |
Source code in zenml/steps/entrypoint_function_utils.py
def validate_reserved_arguments(
signature: inspect.Signature, reserved_arguments: Sequence[str]
) -> None:
"""Validates that the signature does not contain any reserved arguments.
Args:
signature: The signature to validate.
reserved_arguments: The reserved arguments for the signature.
Raises:
RuntimeError: If the signature contains a reserved argument.
"""
for arg in reserved_arguments:
if arg in signature.parameters:
raise RuntimeError(f"Reserved argument name '{arg}'.")
external_artifact
Backward compatibility for the ExternalArtifact
class.
step_decorator
Step decorator function.
step(_func=None, *, name=None, enable_cache=None, enable_artifact_metadata=None, enable_artifact_visualization=None, enable_step_logs=None, experiment_tracker=None, step_operator=None, output_materializers=None, settings=None, extra=None, on_failure=None, on_success=None, model=None)
Outer decorator function for the creation of a ZenML step.
In order to be able to work with parameters such as name
, it features a
nested decorator structure.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
_func |
Optional[~F] |
The decorated function. |
None |
name |
Optional[str] |
The name of the step. If left empty, the name of the decorated function will be used as a fallback. |
None |
enable_cache |
Optional[bool] |
Specify whether caching is enabled for this step. If no value is passed, caching is enabled by default. |
None |
enable_artifact_metadata |
Optional[bool] |
Specify whether metadata is enabled for this step. If no value is passed, metadata is enabled by default. |
None |
enable_artifact_visualization |
Optional[bool] |
Specify whether visualization is enabled for this step. If no value is passed, visualization is enabled by default. |
None |
enable_step_logs |
Optional[bool] |
Specify whether step logs are enabled for this step. |
None |
experiment_tracker |
Optional[str] |
The experiment tracker to use for this step. |
None |
step_operator |
Optional[str] |
The step operator to use for this step. |
None |
output_materializers |
Optional[OutputMaterializersSpecification] |
Output materializers for this step. If given as a dict, the keys must be a subset of the output names of this step. If a single value (type or string) is given, the materializer will be used for all outputs. |
None |
settings |
Optional[Dict[str, SettingsOrDict]] |
Settings for this step. |
None |
extra |
Optional[Dict[str, Any]] |
Extra configurations for this step. |
None |
on_failure |
Optional[HookSpecification] |
Callback function in event of failure of the step. Can be a
function with a single argument of type |
None |
on_success |
Optional[HookSpecification] |
Callback function in event of success of the step. Can be a
function with no arguments, or a source path to such a function
(e.g. |
None |
model |
Optional[Model] |
configuration of the model version in the Model Control Plane. |
None |
Returns:
Type | Description |
---|---|
Union[Type[zenml.steps.base_step.BaseStep], Callable[[~F], Type[zenml.steps.base_step.BaseStep]]] |
The inner decorator which creates the step class based on the ZenML BaseStep |
Source code in zenml/steps/step_decorator.py
def step(
_func: Optional[F] = None,
*,
name: Optional[str] = None,
enable_cache: Optional[bool] = None,
enable_artifact_metadata: Optional[bool] = None,
enable_artifact_visualization: Optional[bool] = None,
enable_step_logs: Optional[bool] = None,
experiment_tracker: Optional[str] = None,
step_operator: Optional[str] = None,
output_materializers: Optional["OutputMaterializersSpecification"] = None,
settings: Optional[Dict[str, "SettingsOrDict"]] = None,
extra: Optional[Dict[str, Any]] = None,
on_failure: Optional["HookSpecification"] = None,
on_success: Optional["HookSpecification"] = None,
model: Optional["Model"] = None,
) -> Union[Type[BaseStep], Callable[[F], Type[BaseStep]]]:
"""Outer decorator function for the creation of a ZenML step.
In order to be able to work with parameters such as `name`, it features a
nested decorator structure.
Args:
_func: The decorated function.
name: The name of the step. If left empty, the name of the decorated
function will be used as a fallback.
enable_cache: Specify whether caching is enabled for this step. If no
value is passed, caching is enabled by default.
enable_artifact_metadata: Specify whether metadata is enabled for this
step. If no value is passed, metadata is enabled by default.
enable_artifact_visualization: Specify whether visualization is enabled
for this step. If no value is passed, visualization is enabled by
default.
enable_step_logs: Specify whether step logs are enabled for this step.
experiment_tracker: The experiment tracker to use for this step.
step_operator: The step operator to use for this step.
output_materializers: Output materializers for this step. If
given as a dict, the keys must be a subset of the output names
of this step. If a single value (type or string) is given, the
materializer will be used for all outputs.
settings: Settings for this step.
extra: Extra configurations for this step.
on_failure: Callback function in event of failure of the step. Can be a
function with a single argument of type `BaseException`, or a source
path to such a function (e.g. `module.my_function`).
on_success: Callback function in event of success of the step. Can be a
function with no arguments, or a source path to such a function
(e.g. `module.my_function`).
model: configuration of the model version in the Model Control Plane.
Returns:
The inner decorator which creates the step class based on the
ZenML BaseStep
"""
def inner_decorator(func: F) -> Type[BaseStep]:
"""Inner decorator function for the creation of a ZenML Step.
Args:
func: types.FunctionType, this function will be used as the
"process" method of the generated Step.
Returns:
The class of a newly generated ZenML Step.
"""
step_name = name or func.__name__
logger.warning(
f"The `@step` decorator that you used to define your {step_name} "
"step is deprecated. Check out the 0.40.0 migration guide for more "
"information on how to migrate your steps to the new syntax: "
"https://docs.zenml.io/reference/migration-guide/migration-zero-forty"
)
return type( # noqa
func.__name__,
(_DecoratedStep,),
{
STEP_INNER_FUNC_NAME: staticmethod(func),
CLASS_CONFIGURATION: {
PARAM_STEP_NAME: name,
PARAM_ENABLE_CACHE: enable_cache,
PARAM_ENABLE_ARTIFACT_METADATA: enable_artifact_metadata,
PARAM_ENABLE_ARTIFACT_VISUALIZATION: enable_artifact_visualization,
PARAM_ENABLE_STEP_LOGS: enable_step_logs,
PARAM_EXPERIMENT_TRACKER: experiment_tracker,
PARAM_STEP_OPERATOR: step_operator,
PARAM_OUTPUT_MATERIALIZERS: output_materializers,
PARAM_SETTINGS: settings,
PARAM_EXTRA_OPTIONS: extra,
PARAM_ON_FAILURE: on_failure,
PARAM_ON_SUCCESS: on_success,
PARAM_MODEL: model,
},
"__module__": func.__module__,
"__doc__": func.__doc__,
},
)
if _func is None:
return inner_decorator
else:
return inner_decorator(_func)
step_environment
Step environment class.
StepEnvironment (BaseEnvironmentComponent)
(Deprecated) Added information about a run inside a step function.
This takes the form of an Environment component. This class can be used from within a pipeline step implementation to access additional information about the runtime parameters of a pipeline step, such as the pipeline name, pipeline run ID and other pipeline runtime information. To use it, access it inside your step function like this:
from zenml.environment import Environment
@step
def my_step(...)
env = Environment().step_environment
do_something_with(env.pipeline_name, env.run_name, env.step_name)
Source code in zenml/steps/step_environment.py
class StepEnvironment(BaseEnvironmentComponent):
"""(Deprecated) Added information about a run inside a step function.
This takes the form of an Environment component. This class can be used from
within a pipeline step implementation to access additional information about
the runtime parameters of a pipeline step, such as the pipeline name,
pipeline run ID and other pipeline runtime information. To use it, access it
inside your step function like this:
```python
from zenml.environment import Environment
@step
def my_step(...)
env = Environment().step_environment
do_something_with(env.pipeline_name, env.run_name, env.step_name)
```
"""
NAME = STEP_ENVIRONMENT_NAME
def __init__(
self,
step_run_info: "StepRunInfo",
cache_enabled: bool,
):
"""Initialize the environment of the currently running step.
Args:
step_run_info: Info about the currently running step.
cache_enabled: Whether caching is enabled for the current step run.
"""
super().__init__()
self._step_run_info = step_run_info
self._cache_enabled = cache_enabled
@property
def pipeline_name(self) -> str:
"""The name of the currently running pipeline.
Returns:
The name of the currently running pipeline.
"""
return self._step_run_info.pipeline.name
@property
def run_name(self) -> str:
"""The name of the current pipeline run.
Returns:
The name of the current pipeline run.
"""
return self._step_run_info.run_name
@property
def step_name(self) -> str:
"""The name of the currently running step.
Returns:
The name of the currently running step.
"""
return self._step_run_info.pipeline_step_name
@property
def step_run_info(self) -> "StepRunInfo":
"""Info about the currently running step.
Returns:
Info about the currently running step.
"""
return self._step_run_info
@property
def cache_enabled(self) -> bool:
"""Returns whether cache is enabled for the step.
Returns:
True if cache is enabled for the step, otherwise False.
"""
return self._cache_enabled
cache_enabled: bool
property
readonly
Returns whether cache is enabled for the step.
Returns:
Type | Description |
---|---|
bool |
True if cache is enabled for the step, otherwise False. |
pipeline_name: str
property
readonly
The name of the currently running pipeline.
Returns:
Type | Description |
---|---|
str |
The name of the currently running pipeline. |
run_name: str
property
readonly
The name of the current pipeline run.
Returns:
Type | Description |
---|---|
str |
The name of the current pipeline run. |
step_name: str
property
readonly
The name of the currently running step.
Returns:
Type | Description |
---|---|
str |
The name of the currently running step. |
step_run_info: StepRunInfo
property
readonly
Info about the currently running step.
Returns:
Type | Description |
---|---|
StepRunInfo |
Info about the currently running step. |
__init__(self, step_run_info, cache_enabled)
special
Initialize the environment of the currently running step.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
step_run_info |
StepRunInfo |
Info about the currently running step. |
required |
cache_enabled |
bool |
Whether caching is enabled for the current step run. |
required |
Source code in zenml/steps/step_environment.py
def __init__(
self,
step_run_info: "StepRunInfo",
cache_enabled: bool,
):
"""Initialize the environment of the currently running step.
Args:
step_run_info: Info about the currently running step.
cache_enabled: Whether caching is enabled for the current step run.
"""
super().__init__()
self._step_run_info = step_run_info
self._cache_enabled = cache_enabled
step_invocation
Step invocation class definition.
StepInvocation
Step invocation class.
Source code in zenml/steps/step_invocation.py
class StepInvocation:
"""Step invocation class."""
def __init__(
self,
id: str,
step: "BaseStep",
input_artifacts: Dict[str, "StepArtifact"],
external_artifacts: Dict[str, "ExternalArtifact"],
model_artifacts_or_metadata: Dict[str, "ModelVersionDataLazyLoader"],
client_lazy_loaders: Dict[str, "ClientLazyLoader"],
parameters: Dict[str, Any],
default_parameters: Dict[str, Any],
upstream_steps: Set[str],
pipeline: "Pipeline",
) -> None:
"""Initialize a step invocation.
Args:
id: The invocation ID.
step: The step that is represented by the invocation.
input_artifacts: The input artifacts for the invocation.
external_artifacts: The external artifacts for the invocation.
model_artifacts_or_metadata: The model artifacts or metadata for
the invocation.
client_lazy_loaders: The client lazy loaders for the invocation.
parameters: The parameters for the invocation.
default_parameters: The default parameters for the invocation.
upstream_steps: The upstream steps for the invocation.
pipeline: The parent pipeline of the invocation.
"""
self.id = id
self.step = step
self.input_artifacts = input_artifacts
self.external_artifacts = external_artifacts
self.model_artifacts_or_metadata = model_artifacts_or_metadata
self.client_lazy_loaders = client_lazy_loaders
self.parameters = parameters
self.default_parameters = default_parameters
self.invocation_upstream_steps = upstream_steps
self.pipeline = pipeline
@property
def upstream_steps(self) -> Set[str]:
"""The upstream steps of the invocation.
Returns:
The upstream steps of the invocation.
"""
return self.invocation_upstream_steps.union(
self._get_and_validate_step_upstream_steps()
)
def _get_and_validate_step_upstream_steps(self) -> Set[str]:
"""Validates the upstream steps defined on the step instance.
This is only allowed in legacy pipelines when calling `step.after(...)`
and we need to make sure that both the upstream and downstream steps
of such a relationship are only invoked once inside a pipeline.
Returns:
The upstream steps defined on the step instance.
"""
def _verify_single_invocation(step: "BaseStep") -> str:
invocations = {
invocation
for invocation in self.pipeline.invocations.values()
if invocation.step is step
}
if len(invocations) > 1:
raise RuntimeError(
"Setting upstream steps for a step using "
"`step_1.after(step_2)` is not allowed in combination "
"with calling one of the two steps multiple times."
)
return invocations.pop().id
if self.step.upstream_steps:
# If the step has upstream steps, make sure it only got invoked once
_verify_single_invocation(step=self.step)
upstream_steps = set()
for upstream_step in self.step.upstream_steps:
upstream_step_invocation_id = _verify_single_invocation(
step=upstream_step
)
upstream_steps.add(upstream_step_invocation_id)
return upstream_steps
def finalize(self, parameters_to_ignore: Set[str]) -> "StepConfiguration":
"""Finalizes a step invocation.
It will validate the upstream steps and run final configurations on the
step that is represented by the invocation.
Args:
parameters_to_ignore: Set of parameters that should not be applied
to the step instance.
Returns:
The finalized step configuration.
"""
from zenml.artifacts.external_artifact_config import (
ExternalArtifactConfiguration,
)
# Validate the upstream steps for legacy .after() calls
self._get_and_validate_step_upstream_steps()
parameters_to_apply = {
key: value
for key, value in self.parameters.items()
if key not in parameters_to_ignore
}
parameters_to_apply.update(
{
key: value
for key, value in self.default_parameters.items()
if key not in parameters_to_ignore
and key not in parameters_to_apply
}
)
self.step.configure(parameters=parameters_to_apply)
external_artifacts: Dict[str, ExternalArtifactConfiguration] = {}
for key, artifact in self.external_artifacts.items():
if artifact.value is not None:
artifact.upload_by_value()
external_artifacts[key] = artifact.config
return self.step._finalize_configuration(
input_artifacts=self.input_artifacts,
external_artifacts=external_artifacts,
model_artifacts_or_metadata=self.model_artifacts_or_metadata,
client_lazy_loaders=self.client_lazy_loaders,
)
upstream_steps: Set[str]
property
readonly
The upstream steps of the invocation.
Returns:
Type | Description |
---|---|
Set[str] |
The upstream steps of the invocation. |
__init__(self, id, step, input_artifacts, external_artifacts, model_artifacts_or_metadata, client_lazy_loaders, parameters, default_parameters, upstream_steps, pipeline)
special
Initialize a step invocation.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
id |
str |
The invocation ID. |
required |
step |
BaseStep |
The step that is represented by the invocation. |
required |
input_artifacts |
Dict[str, StepArtifact] |
The input artifacts for the invocation. |
required |
external_artifacts |
Dict[str, ExternalArtifact] |
The external artifacts for the invocation. |
required |
model_artifacts_or_metadata |
Dict[str, ModelVersionDataLazyLoader] |
The model artifacts or metadata for the invocation. |
required |
client_lazy_loaders |
Dict[str, ClientLazyLoader] |
The client lazy loaders for the invocation. |
required |
parameters |
Dict[str, Any] |
The parameters for the invocation. |
required |
default_parameters |
Dict[str, Any] |
The default parameters for the invocation. |
required |
upstream_steps |
Set[str] |
The upstream steps for the invocation. |
required |
pipeline |
Pipeline |
The parent pipeline of the invocation. |
required |
Source code in zenml/steps/step_invocation.py
def __init__(
self,
id: str,
step: "BaseStep",
input_artifacts: Dict[str, "StepArtifact"],
external_artifacts: Dict[str, "ExternalArtifact"],
model_artifacts_or_metadata: Dict[str, "ModelVersionDataLazyLoader"],
client_lazy_loaders: Dict[str, "ClientLazyLoader"],
parameters: Dict[str, Any],
default_parameters: Dict[str, Any],
upstream_steps: Set[str],
pipeline: "Pipeline",
) -> None:
"""Initialize a step invocation.
Args:
id: The invocation ID.
step: The step that is represented by the invocation.
input_artifacts: The input artifacts for the invocation.
external_artifacts: The external artifacts for the invocation.
model_artifacts_or_metadata: The model artifacts or metadata for
the invocation.
client_lazy_loaders: The client lazy loaders for the invocation.
parameters: The parameters for the invocation.
default_parameters: The default parameters for the invocation.
upstream_steps: The upstream steps for the invocation.
pipeline: The parent pipeline of the invocation.
"""
self.id = id
self.step = step
self.input_artifacts = input_artifacts
self.external_artifacts = external_artifacts
self.model_artifacts_or_metadata = model_artifacts_or_metadata
self.client_lazy_loaders = client_lazy_loaders
self.parameters = parameters
self.default_parameters = default_parameters
self.invocation_upstream_steps = upstream_steps
self.pipeline = pipeline
finalize(self, parameters_to_ignore)
Finalizes a step invocation.
It will validate the upstream steps and run final configurations on the step that is represented by the invocation.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
parameters_to_ignore |
Set[str] |
Set of parameters that should not be applied to the step instance. |
required |
Returns:
Type | Description |
---|---|
StepConfiguration |
The finalized step configuration. |
Source code in zenml/steps/step_invocation.py
def finalize(self, parameters_to_ignore: Set[str]) -> "StepConfiguration":
"""Finalizes a step invocation.
It will validate the upstream steps and run final configurations on the
step that is represented by the invocation.
Args:
parameters_to_ignore: Set of parameters that should not be applied
to the step instance.
Returns:
The finalized step configuration.
"""
from zenml.artifacts.external_artifact_config import (
ExternalArtifactConfiguration,
)
# Validate the upstream steps for legacy .after() calls
self._get_and_validate_step_upstream_steps()
parameters_to_apply = {
key: value
for key, value in self.parameters.items()
if key not in parameters_to_ignore
}
parameters_to_apply.update(
{
key: value
for key, value in self.default_parameters.items()
if key not in parameters_to_ignore
and key not in parameters_to_apply
}
)
self.step.configure(parameters=parameters_to_apply)
external_artifacts: Dict[str, ExternalArtifactConfiguration] = {}
for key, artifact in self.external_artifacts.items():
if artifact.value is not None:
artifact.upload_by_value()
external_artifacts[key] = artifact.config
return self.step._finalize_configuration(
input_artifacts=self.input_artifacts,
external_artifacts=external_artifacts,
model_artifacts_or_metadata=self.model_artifacts_or_metadata,
client_lazy_loaders=self.client_lazy_loaders,
)
step_output
Step output class.
Output
A named tuple with a default name that cannot be overridden.
Source code in zenml/steps/step_output.py
class Output(object):
"""A named tuple with a default name that cannot be overridden."""
def __init__(self, **kwargs: Type[Any]):
"""Initializes the output.
Args:
**kwargs: The output values.
"""
self.outputs = NamedTuple("ZenOutput", **kwargs) # type: ignore[misc]
def items(self) -> Iterator[Tuple[str, Type[Any]]]:
"""Yields a tuple of type (output_name, output_type).
Yields:
A tuple of type (output_name, output_type).
"""
yield from self.outputs.__annotations__.items()
__init__(self, **kwargs)
special
Initializes the output.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
**kwargs |
Type[Any] |
The output values. |
{} |
Source code in zenml/steps/step_output.py
def __init__(self, **kwargs: Type[Any]):
"""Initializes the output.
Args:
**kwargs: The output values.
"""
self.outputs = NamedTuple("ZenOutput", **kwargs) # type: ignore[misc]
items(self)
Yields a tuple of type (output_name, output_type).
Yields:
Type | Description |
---|---|
Iterator[Tuple[str, Type[Any]]] |
A tuple of type (output_name, output_type). |
Source code in zenml/steps/step_output.py
def items(self) -> Iterator[Tuple[str, Type[Any]]]:
"""Yields a tuple of type (output_name, output_type).
Yields:
A tuple of type (output_name, output_type).
"""
yield from self.outputs.__annotations__.items()
utils
Utility functions and classes to run ZenML steps.
OnlyNoneReturnsVisitor (ReturnVisitor)
Checks whether a function AST contains only None
returns.
Source code in zenml/steps/utils.py
class OnlyNoneReturnsVisitor(ReturnVisitor):
"""Checks whether a function AST contains only `None` returns."""
def __init__(self) -> None:
"""Initializes a visitor instance."""
super().__init__()
self.has_only_none_returns = True
def visit_Return(self, node: ast.Return) -> None:
"""Visit a return statement.
Args:
node: The return statement to visit.
"""
if node.value is not None:
if isinstance(node.value, (ast.Constant, ast.NameConstant)):
if node.value.value is None:
return
self.has_only_none_returns = False
__init__(self)
special
Initializes a visitor instance.
Source code in zenml/steps/utils.py
def __init__(self) -> None:
"""Initializes a visitor instance."""
super().__init__()
self.has_only_none_returns = True
visit_Return(self, node)
Visit a return statement.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
node |
Return |
The return statement to visit. |
required |
Source code in zenml/steps/utils.py
def visit_Return(self, node: ast.Return) -> None:
"""Visit a return statement.
Args:
node: The return statement to visit.
"""
if node.value is not None:
if isinstance(node.value, (ast.Constant, ast.NameConstant)):
if node.value.value is None:
return
self.has_only_none_returns = False
OutputSignature (BaseModel)
The signature of an output artifact.
Source code in zenml/steps/utils.py
class OutputSignature(BaseModel):
"""The signature of an output artifact."""
resolved_annotation: Any = None
artifact_config: Optional[ArtifactConfig] = None
has_custom_name: bool = False
def get_output_types(self) -> Tuple[Any, ...]:
"""Get all output types that match the type annotation.
Returns:
All output types that match the type annotation.
"""
if self.resolved_annotation is Any:
return ()
if typing_utils.is_union(
typing_utils.get_origin(self.resolved_annotation)
or self.resolved_annotation
):
return tuple(
type(None)
if typing_utils.is_none_type(output_type)
else output_type
for output_type in get_args(self.resolved_annotation)
)
else:
return (self.resolved_annotation,)
get_output_types(self)
Get all output types that match the type annotation.
Returns:
Type | Description |
---|---|
Tuple[Any, ...] |
All output types that match the type annotation. |
Source code in zenml/steps/utils.py
def get_output_types(self) -> Tuple[Any, ...]:
"""Get all output types that match the type annotation.
Returns:
All output types that match the type annotation.
"""
if self.resolved_annotation is Any:
return ()
if typing_utils.is_union(
typing_utils.get_origin(self.resolved_annotation)
or self.resolved_annotation
):
return tuple(
type(None)
if typing_utils.is_none_type(output_type)
else output_type
for output_type in get_args(self.resolved_annotation)
)
else:
return (self.resolved_annotation,)
ReturnVisitor (NodeVisitor)
AST visitor class that can be subclassed to visit function returns.
Source code in zenml/steps/utils.py
class ReturnVisitor(ast.NodeVisitor):
"""AST visitor class that can be subclassed to visit function returns."""
def __init__(self, ignore_nested_functions: bool = True) -> None:
"""Initializes a return visitor instance.
Args:
ignore_nested_functions: If `True`, will skip visiting nested
functions.
"""
self._ignore_nested_functions = ignore_nested_functions
self._inside_function = False
def _visit_function(
self, node: Union[ast.FunctionDef, ast.AsyncFunctionDef]
) -> None:
"""Visit a (async) function definition node.
Args:
node: The node to visit.
"""
if self._ignore_nested_functions and self._inside_function:
# We're already inside a function definition and should ignore
# nested functions so we don't want to recurse any further
return
self._inside_function = True
self.generic_visit(node)
visit_FunctionDef = _visit_function
visit_AsyncFunctionDef = _visit_function
__init__(self, ignore_nested_functions=True)
special
Initializes a return visitor instance.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
ignore_nested_functions |
bool |
If |
True |
Source code in zenml/steps/utils.py
def __init__(self, ignore_nested_functions: bool = True) -> None:
"""Initializes a return visitor instance.
Args:
ignore_nested_functions: If `True`, will skip visiting nested
functions.
"""
self._ignore_nested_functions = ignore_nested_functions
self._inside_function = False
visit_AsyncFunctionDef(self, node)
Visit a (async) function definition node.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
node |
Union[ast.FunctionDef, ast.AsyncFunctionDef] |
The node to visit. |
required |
Source code in zenml/steps/utils.py
def _visit_function(
self, node: Union[ast.FunctionDef, ast.AsyncFunctionDef]
) -> None:
"""Visit a (async) function definition node.
Args:
node: The node to visit.
"""
if self._ignore_nested_functions and self._inside_function:
# We're already inside a function definition and should ignore
# nested functions so we don't want to recurse any further
return
self._inside_function = True
self.generic_visit(node)
visit_FunctionDef(self, node)
Visit a (async) function definition node.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
node |
Union[ast.FunctionDef, ast.AsyncFunctionDef] |
The node to visit. |
required |
Source code in zenml/steps/utils.py
def _visit_function(
self, node: Union[ast.FunctionDef, ast.AsyncFunctionDef]
) -> None:
"""Visit a (async) function definition node.
Args:
node: The node to visit.
"""
if self._ignore_nested_functions and self._inside_function:
# We're already inside a function definition and should ignore
# nested functions so we don't want to recurse any further
return
self._inside_function = True
self.generic_visit(node)
TupleReturnVisitor (ReturnVisitor)
Checks whether a function AST contains tuple returns.
Source code in zenml/steps/utils.py
class TupleReturnVisitor(ReturnVisitor):
"""Checks whether a function AST contains tuple returns."""
def __init__(self) -> None:
"""Initializes a visitor instance."""
super().__init__()
self.has_tuple_return = False
def visit_Return(self, node: ast.Return) -> None:
"""Visit a return statement.
Args:
node: The return statement to visit.
"""
if isinstance(node.value, ast.Tuple) and len(node.value.elts) > 1:
self.has_tuple_return = True
__init__(self)
special
Initializes a visitor instance.
Source code in zenml/steps/utils.py
def __init__(self) -> None:
"""Initializes a visitor instance."""
super().__init__()
self.has_tuple_return = False
visit_Return(self, node)
Visit a return statement.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
node |
Return |
The return statement to visit. |
required |
Source code in zenml/steps/utils.py
def visit_Return(self, node: ast.Return) -> None:
"""Visit a return statement.
Args:
node: The return statement to visit.
"""
if isinstance(node.value, ast.Tuple) and len(node.value.elts) > 1:
self.has_tuple_return = True
get_args(obj)
Get arguments of a type annotation.
Examples:
get_args(Union[int, str]) == (int, str)
Parameters:
Name | Type | Description | Default |
---|---|---|---|
obj |
Any |
The annotation. |
required |
Returns:
Type | Description |
---|---|
Tuple[Any, ...] |
The args of the annotation. |
Source code in zenml/steps/utils.py
def get_args(obj: Any) -> Tuple[Any, ...]:
"""Get arguments of a type annotation.
Example:
`get_args(Union[int, str]) == (int, str)`
Args:
obj: The annotation.
Returns:
The args of the annotation.
"""
return tuple(
typing_utils.get_origin(v) or v for v in typing_utils.get_args(obj)
)
get_artifact_config_from_annotation_metadata(annotation)
Get the artifact config from the annotation metadata of a step output.
Examples:
get_output_name_from_annotation_metadata(int) # None
get_output_name_from_annotation_metadata(Annotated[int, "name"] # ArtifactConfig(name="name")
get_output_name_from_annotation_metadata(Annotated[int, ArtifactConfig(name="name", model_name="foo")] # ArtifactConfig(name="name", model_name="foo")
Parameters:
Name | Type | Description | Default |
---|---|---|---|
annotation |
Any |
The type annotation. |
required |
Exceptions:
Type | Description |
---|---|
ValueError |
If the annotation is not following the expected format or if the name was specified multiple times or is an empty string. |
Returns:
Type | Description |
---|---|
Optional[zenml.artifacts.artifact_config.ArtifactConfig] |
The artifact config. |
Source code in zenml/steps/utils.py
def get_artifact_config_from_annotation_metadata(
annotation: Any,
) -> Optional[ArtifactConfig]:
"""Get the artifact config from the annotation metadata of a step output.
Example:
```python
get_output_name_from_annotation_metadata(int) # None
get_output_name_from_annotation_metadata(Annotated[int, "name"] # ArtifactConfig(name="name")
get_output_name_from_annotation_metadata(Annotated[int, ArtifactConfig(name="name", model_name="foo")] # ArtifactConfig(name="name", model_name="foo")
```
Args:
annotation: The type annotation.
Raises:
ValueError: If the annotation is not following the expected format
or if the name was specified multiple times or is an empty string.
Returns:
The artifact config.
"""
if (typing_utils.get_origin(annotation) or annotation) is not Annotated:
return None
annotation, *metadata = typing_utils.get_args(annotation)
error_message = (
"Artifact annotation should only contain two elements: the artifact "
"type, and either an output name or an `ArtifactConfig`, e.g.: "
"`Annotated[int, 'output_name']` or "
"`Annotated[int, ArtifactConfig(name='output_name'), ...]`."
)
if len(metadata) > 2:
raise ValueError(error_message)
# Loop over all values to also support legacy annotations of the form
# `Annotated[int, 'output_name', ArtifactConfig(...)]`
output_name = None
artifact_config = None
for metadata_instance in metadata:
if isinstance(metadata_instance, str):
if output_name is not None:
raise ValueError(error_message)
output_name = metadata_instance
elif isinstance(metadata_instance, ArtifactConfig):
if artifact_config is not None:
raise ValueError(error_message)
artifact_config = metadata_instance
else:
raise ValueError(error_message)
# Consolidate output name
if artifact_config and artifact_config.name:
if output_name is not None:
raise ValueError(error_message)
elif output_name:
if not artifact_config:
artifact_config = ArtifactConfig(name=output_name)
elif not artifact_config.name:
artifact_config = artifact_config.model_copy()
artifact_config.name = output_name
if artifact_config and artifact_config.name == "":
raise ValueError("Output name cannot be an empty string.")
return artifact_config
has_only_none_returns(func)
Checks whether a function contains only None
returns.
A None
return could be either an explicit return None
or an empty
return
statement.
Examples:
def f1():
return None
def f2():
return
def f3(condition):
if condition:
return None
else:
return 1
has_only_none_returns(f1) # True
has_only_none_returns(f2) # True
has_only_none_returns(f3) # False
Parameters:
Name | Type | Description | Default |
---|---|---|---|
func |
Callable[..., Any] |
The function to check. |
required |
Returns:
Type | Description |
---|---|
bool |
Whether the function contains only |
Source code in zenml/steps/utils.py
def has_only_none_returns(func: Callable[..., Any]) -> bool:
"""Checks whether a function contains only `None` returns.
A `None` return could be either an explicit `return None` or an empty
`return` statement.
Example:
```python
def f1():
return None
def f2():
return
def f3(condition):
if condition:
return None
else:
return 1
has_only_none_returns(f1) # True
has_only_none_returns(f2) # True
has_only_none_returns(f3) # False
```
Args:
func: The function to check.
Returns:
Whether the function contains only `None` returns.
"""
source = textwrap.dedent(source_code_utils.get_source_code(func))
tree = ast.parse(source)
visitor = OnlyNoneReturnsVisitor()
visitor.visit(tree)
return visitor.has_only_none_returns
has_tuple_return(func)
Checks whether a function returns multiple values.
Multiple values means that the return
statement is followed by a tuple
(with or without brackets).
Examples:
def f1():
return 1, 2
def f2():
return (1, 2)
def f3():
var = (1, 2)
return var
has_tuple_return(f1) # True
has_tuple_return(f2) # True
has_tuple_return(f3) # False
Parameters:
Name | Type | Description | Default |
---|---|---|---|
func |
Callable[..., Any] |
The function to check. |
required |
Returns:
Type | Description |
---|---|
bool |
Whether the function returns multiple values. |
Source code in zenml/steps/utils.py
def has_tuple_return(func: Callable[..., Any]) -> bool:
"""Checks whether a function returns multiple values.
Multiple values means that the `return` statement is followed by a tuple
(with or without brackets).
Example:
```python
def f1():
return 1, 2
def f2():
return (1, 2)
def f3():
var = (1, 2)
return var
has_tuple_return(f1) # True
has_tuple_return(f2) # True
has_tuple_return(f3) # False
```
Args:
func: The function to check.
Returns:
Whether the function returns multiple values.
"""
source = textwrap.dedent(source_code_utils.get_source_code(func))
tree = ast.parse(source)
visitor = TupleReturnVisitor()
visitor.visit(tree)
return visitor.has_tuple_return
log_step_metadata(metadata, step_name=None, pipeline_name_id_or_prefix=None, run_id=None)
Logs step metadata.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
metadata |
Dict[str, MetadataType] |
The metadata to log. |
required |
step_name |
Optional[str] |
The name of the step to log metadata for. Can be omitted when being called inside a step. |
None |
pipeline_name_id_or_prefix |
Union[uuid.UUID, str] |
The name of the pipeline to log metadata for. Can be omitted when being called inside a step. |
None |
run_id |
Optional[str] |
The ID of the run to log metadata for. Can be omitted when being called inside a step. |
None |
Exceptions:
Type | Description |
---|---|
ValueError |
If no step name is provided and the function is not called from within a step or if no pipeline name or ID is provided and the function is not called from within a step. |
Source code in zenml/steps/utils.py
def log_step_metadata(
metadata: Dict[str, "MetadataType"],
step_name: Optional[str] = None,
pipeline_name_id_or_prefix: Optional[Union[str, UUID]] = None,
run_id: Optional[str] = None,
) -> None:
"""Logs step metadata.
Args:
metadata: The metadata to log.
step_name: The name of the step to log metadata for. Can be omitted
when being called inside a step.
pipeline_name_id_or_prefix: The name of the pipeline to log metadata
for. Can be omitted when being called inside a step.
run_id: The ID of the run to log metadata for. Can be omitted when
being called inside a step.
Raises:
ValueError: If no step name is provided and the function is not called
from within a step or if no pipeline name or ID is provided and
the function is not called from within a step.
"""
step_context = None
if not step_name:
with contextlib.suppress(RuntimeError):
step_context = get_step_context()
step_name = step_context.step_name
# not running within a step and no user-provided step name
if not step_name:
raise ValueError(
"No step name provided and you are not running "
"within a step. Please provide a step name."
)
client = Client()
if step_context:
step_run_id = step_context.step_run.id
elif run_id:
step_run_id = UUID(int=int(run_id))
else:
if not pipeline_name_id_or_prefix:
raise ValueError(
"No pipeline name or ID provided and you are not running "
"within a step. Please provide a pipeline name or ID, or "
"provide a run ID."
)
pipeline_run = client.get_pipeline(
name_id_or_prefix=pipeline_name_id_or_prefix,
).last_run
step_run_id = pipeline_run.steps[step_name].id
client.create_run_metadata(
metadata=metadata,
resource_id=step_run_id,
resource_type=MetadataResourceTypes.STEP_RUN,
)
parse_return_type_annotations(func, enforce_type_annotations=False)
Parse the return type annotation of a step function.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
func |
Callable[..., Any] |
The step function. |
required |
enforce_type_annotations |
bool |
If |
False |
Exceptions:
Type | Description |
---|---|
RuntimeError |
If the output annotation has variable length or contains duplicate output names. |
RuntimeError |
If type annotations should be enforced and a type annotation is missing. |
Returns:
Type | Description |
---|---|
Dict[str, zenml.steps.utils.OutputSignature] |
|
Source code in zenml/steps/utils.py
def parse_return_type_annotations(
func: Callable[..., Any], enforce_type_annotations: bool = False
) -> Dict[str, OutputSignature]:
"""Parse the return type annotation of a step function.
Args:
func: The step function.
enforce_type_annotations: If `True`, raises an exception if a type
annotation is missing.
Raises:
RuntimeError: If the output annotation has variable length or contains
duplicate output names.
RuntimeError: If type annotations should be enforced and a type
annotation is missing.
Returns:
- A dictionary mapping output names to their output signatures.
"""
signature = inspect.signature(func, follow_wrapped=True)
return_annotation = signature.return_annotation
output_name: Optional[str]
# Return type annotated as `None`
if return_annotation is None:
return {}
# Return type not annotated -> check whether `None` or `Any` should be used
if return_annotation is signature.empty:
if enforce_type_annotations:
raise RuntimeError(
"Missing return type annotation for step function "
f"'{func.__name__}'."
)
elif has_only_none_returns(func):
return {}
else:
return_annotation = Any
# Return type annotated using deprecated `Output(...)`
if isinstance(return_annotation, Output):
logger.warning(
"Using the `Output` class to define the outputs of your steps is "
"deprecated. You should instead use the standard Python way of "
"type annotating your functions. Check out our documentation "
"https://docs.zenml.io/how-to/build-pipelines/step-output-typing-and-annotation "
"for more information on how to assign custom names to your step "
"outputs."
)
return {
output_name: OutputSignature(
resolved_annotation=resolve_type_annotation(output_type),
artifact_config=None,
has_custom_name=True,
)
for output_name, output_type in return_annotation.items()
}
elif typing_utils.get_origin(return_annotation) is tuple:
requires_multiple_artifacts = has_tuple_return(func)
if requires_multiple_artifacts:
output_signature: Dict[str, Any] = {}
args = typing_utils.get_args(return_annotation)
if args[-1] is Ellipsis:
raise RuntimeError(
"Variable length output annotations are not allowed."
)
for i, annotation in enumerate(args):
resolved_annotation = resolve_type_annotation(annotation)
artifact_config = get_artifact_config_from_annotation_metadata(
annotation
)
output_name = artifact_config.name if artifact_config else None
has_custom_name = output_name is not None
output_name = output_name or f"output_{i}"
if output_name in output_signature:
raise RuntimeError(f"Duplicate output name {output_name}.")
output_signature[output_name] = OutputSignature(
resolved_annotation=resolved_annotation,
artifact_config=artifact_config,
has_custom_name=has_custom_name,
)
return output_signature
# Return type is annotated as single value or is a tuple
resolved_annotation = resolve_type_annotation(return_annotation)
artifact_config = get_artifact_config_from_annotation_metadata(
return_annotation
)
output_name = artifact_config.name if artifact_config else None
has_custom_name = output_name is not None
output_name = output_name or SINGLE_RETURN_OUT_NAME
return {
output_name: OutputSignature(
resolved_annotation=resolved_annotation,
artifact_config=artifact_config,
has_custom_name=has_custom_name,
)
}
resolve_type_annotation(obj)
Returns the non-generic class for generic aliases of the typing module.
Example: if the input object is typing.Dict
, this method will return the
concrete class dict
.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
obj |
Any |
The object to resolve. |
required |
Returns:
Type | Description |
---|---|
Any |
The non-generic class for generic aliases of the typing module. |
Source code in zenml/steps/utils.py
def resolve_type_annotation(obj: Any) -> Any:
"""Returns the non-generic class for generic aliases of the typing module.
Example: if the input object is `typing.Dict`, this method will return the
concrete class `dict`.
Args:
obj: The object to resolve.
Returns:
The non-generic class for generic aliases of the typing module.
"""
origin = typing_utils.get_origin(obj) or obj
if origin is Annotated:
annotation, *_ = typing_utils.get_args(obj)
return resolve_type_annotation(annotation)
elif typing_utils.is_union(origin):
return obj
return origin
run_as_single_step_pipeline(__step, *args, **kwargs)
Runs the step as a single step pipeline.
- All inputs that are not JSON serializable will be uploaded to the artifact store before the pipeline is being executed.
- All output artifacts of the step will be loaded using the materializer that was used to store them.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
*args |
Any |
Entrypoint function arguments. |
() |
**kwargs |
Any |
Entrypoint function keyword arguments. |
{} |
Exceptions:
Type | Description |
---|---|
RuntimeError |
If the step execution failed. |
StepInterfaceError |
If the arguments to the entrypoint function are invalid. |
Returns:
Type | Description |
---|---|
Any |
The output of the step entrypoint function. |
Source code in zenml/steps/utils.py
def run_as_single_step_pipeline(
__step: "BaseStep", *args: Any, **kwargs: Any
) -> Any:
"""Runs the step as a single step pipeline.
- All inputs that are not JSON serializable will be uploaded to the
artifact store before the pipeline is being executed.
- All output artifacts of the step will be loaded using the materializer
that was used to store them.
Args:
*args: Entrypoint function arguments.
**kwargs: Entrypoint function keyword arguments.
Raises:
RuntimeError: If the step execution failed.
StepInterfaceError: If the arguments to the entrypoint function are
invalid.
Returns:
The output of the step entrypoint function.
"""
from zenml import ExternalArtifact, pipeline
from zenml.config.base_settings import BaseSettings
from zenml.new.pipelines.run_utils import (
wait_for_pipeline_run_to_finish,
)
logger.info(
"Running single step pipeline to execute step `%s`", __step.name
)
try:
validated_arguments = (
inspect.signature(__step.entrypoint)
.bind(*args, **kwargs)
.arguments
)
except TypeError as e:
raise StepInterfaceError(
"Invalid step function entrypoint arguments. Check out the "
"error above for more details."
) from e
inputs: Dict[str, Any] = {}
for key, value in validated_arguments.items():
try:
__step.entrypoint_definition.validate_input(key=key, value=value)
inputs[key] = value
except Exception:
inputs[key] = ExternalArtifact(value=value)
orchestrator = Client().active_stack.orchestrator
pipeline_settings: Any = {}
if "synchronous" in orchestrator.config.model_fields:
# Make sure the orchestrator runs sync so we stream the logs
key = settings_utils.get_stack_component_setting_key(orchestrator)
pipeline_settings[key] = BaseSettings(synchronous=True)
@pipeline(name=__step.name, enable_cache=False, settings=pipeline_settings)
def single_step_pipeline() -> None:
__step(**inputs)
run = single_step_pipeline.with_options(unlisted=True)()
run = wait_for_pipeline_run_to_finish(run.id)
if run.status != ExecutionStatus.COMPLETED:
raise RuntimeError("Failed to execute step %s.", __step.name)
# 4. Load output artifacts
step_run = next(iter(run.steps.values()))
outputs = [
step_run.outputs[output_name].load()
for output_name in step_run.config.outputs.keys()
]
if len(outputs) == 0:
return None
elif len(outputs) == 1:
return outputs[0]
else:
return tuple(outputs)