Share via

AutoMLRun Class

  • Reference

Represents an automated ML experiment run in Azure Machine Learning.

The AutoMLRun class can be used to manage a run, check run status, and retrieve run details once an AutoML run is submitted. For more information on working with experiment runs, see the Run class.

Initialize an AutoML run.

Constructor

AutoMLRun(experiment, run_id, **kwargs)

Parameters

Name Description
experiment
Required
Experiment

The experiment associated with the run.
run_id
Required
str

The ID of the run.
experiment
Required
Experiment

The experiment associated with the run.
run_id
Required
str

The ID of the run.

Remarks

An AutoMLRun object is returned when you use the submit method of an experiment.

To retrieve a run that has already started, use the following code:


   from azureml.train.automl.run import AutoMLRun
   ws = Workspace.from_config()
   experiment = ws.experiments['my-experiment-name']
   automl_run = AutoMLRun(experiment, run_id = 'AutoML_9fe201fe-89fd-41cc-905f-2f41a5a98883')

Methods

cancel

Cancel an AutoML run.

Return True if the AutoML run was canceled successfully.
cancel_iteration

Cancel a particular child run.
complete

Complete an AutoML Run.
continue_experiment

Continue an existing AutoML experiment.
fail

Fail an AutoML Run.

Optionally set the Error property of the run with a message or exception passed to error_details.
get_best_child

Return the child run with the best score for this AutoML Run.
get_guardrails

Print and return detailed results from running Guardrail verification.
get_output

Return the run with the corresponding best pipeline that has already been tested.

If no input parameters are provided, get_output returns the best pipeline according to the primary metric. Alternatively, you can use either the iteration or metric parameter to retrieve a particular iteration or the best run per provided metric, respectively.
get_run_sdk_dependencies

Get the SDK run dependencies for a given run.
pause

Return True if the AutoML run was paused successfully.

This method is not implemented.
register_model

Register the model with AzureML ACI service.
resume

Return True if the AutoML run was resumed successfully.

This method is not implemented.
retry

Return True if the AutoML run was retried successfully.

This method is not implemented.
summary

Get a table containing a summary of algorithms attempted and their scores.
wait_for_completion

Wait for the completion of this run.

Returns the status object after the wait.

cancel

Cancel an AutoML run.

Return True if the AutoML run was canceled successfully.

cancel()

Returns

Type Description

None

cancel_iteration

Cancel a particular child run.

cancel_iteration(iteration)

Parameters

Name Description
iteration
Required
int

The iteration to cancel.

Returns

Type Description

None

complete

Complete an AutoML Run.

complete(**kwargs)

Returns

Type Description

None

continue_experiment

Continue an existing AutoML experiment.

continue_experiment(X=None, y=None, sample_weight=None, X_valid=None, y_valid=None, sample_weight_valid=None, data=None, label=None, columns=None, cv_splits_indices=None, spark_context=None, experiment_timeout_hours=None, experiment_exit_score=None, iterations=None, show_output=False, training_data=None, validation_data=None, **kwargs)

Parameters

Name Description
X
DataFrame or ndarray or <xref:azureml.dataprep.Dataflow>

Training features.

Default value: None
y
DataFrame or ndarray or <xref:azureml.dataprep.Dataflow>

Training labels.

Default value: None
sample_weight
DataFrame or ndarray or <xref:azureml.dataprep.Dataflow>

Sample weights for training data.

Default value: None
X_valid
DataFrame or ndarray or <xref:azureml.dataprep.Dataflow>

Validation features.

Default value: None
y_valid
DataFrame or ndarray or <xref:azureml.dataprep.Dataflow>

Validation labels.

Default value: None
sample_weight_valid
DataFrame or ndarray or <xref:azureml.dataprep.Dataflow>

validation set sample weights.

Default value: None
data
DataFrame

Training features and label.

Default value: None
label
str

Label column in data.

Default value: None
columns
list(str)

A list of allowed columns in the data to use as features.

Default value: None
cv_splits_indices
ndarray

Indices where to split training data for cross validation. Each row is a separate cross fold and within each crossfold, provide 2 arrays, the first with the indices for samples to use for training data and the second with the indices to use for validation data. i.e [[t1, v1], [t2, v2], ...] where t1 is the training indices for the first cross fold and v1 is the validation indices for the first cross fold.

Default value: None
spark_context
<xref:SparkContext>

Spark context, only applicable when used inside azure databricks/spark environment.

Default value: None
experiment_timeout_hours
float

How many additional hours to run this experiment for.

Default value: None
experiment_exit_score
int

If specified indicates that the experiment is terminated when this value is reached.

Default value: None
iterations
int

How many additional iterations to run for this experiment.

Default value: None
show_output
bool

Flag indicating whether to print output to console.

Default value: False
training_data
<xref:azureml.dataprep.Dataflow> or DataFrame

Input training data.

Default value: None
validation_data
<xref:azureml.dataprep.Dataflow> or DataFrame

Validation data.

Default value: None

Returns

Type Description
AutoMLRun

The AutoML parent run.

Exceptions

Type Description
ConfigException
InvalidRunState

fail

Fail an AutoML Run.

Optionally set the Error property of the run with a message or exception passed to error_details.

fail(error_details=None, error_code=None, _set_status=True, **kwargs)

Parameters

Name Description
error_details
str or BaseException

Optional details of the error.

Default value: None
error_code
str

Optional error code of the error for the error classification.

Default value: None
_set_status
bool

Indicates whether to send the status event for tracking.

Default value: True

get_best_child

Return the child run with the best score for this AutoML Run.

get_best_child(metric: str | None = None, onnx_compatible: bool = False, **kwargs: Any) -> Run

Parameters

Name Description
metric
str

The metric to use to when selecting the best run to return. Defaults to the primary metric.

Default value: None
onnx_compatible

Whether to only return runs that generated onnx models.

Default value: False
kwargs
Required

Returns

Type Description

AutoML Child Run.

get_guardrails

Print and return detailed results from running Guardrail verification.

get_guardrails(to_console: bool = True) -> Dict[str, Any]

Parameters

Name Description
to_console
bool

Indicates whether to write the verification results to the console.

Default value: True

Returns

Type Description
dict

A dictionary of verifier results.

Exceptions

Type Description
UserErrorException

get_output

Return the run with the corresponding best pipeline that has already been tested.

If no input parameters are provided, get_output returns the best pipeline according to the primary metric. Alternatively, you can use either the iteration or metric parameter to retrieve a particular iteration or the best run per provided metric, respectively.

get_output(iteration: int | None = None, metric: str | None = None, return_onnx_model: bool = False, return_split_onnx_model: SplitOnnxModelName | None = None, **kwargs: Any) -> Tuple[Run, Any]

Parameters

Name Description
iteration
int

The iteration number of the corresponding run and fitted model to return.

Default value: None
metric
str

The metric to use to when selecting the best run and fitted model to return.

Default value: None
return_onnx_model
bool

This method will return the converted ONNX model if the enable_onnx_compatible_models parameter was set to True in the AutoMLConfig object.

Default value: False
return_split_onnx_model
SplitOnnxModelName

The type of the split onnx model to return

Default value: None

Returns

Type Description
Run, <xref:Model>

The run, the corresponding fitted model.

Exceptions

Type Description
ConfigException
OnnxConvertException
NotFoundException

Remarks

If you'd like to inspect the preprocessor(s) and algorithm (estimator) used, you can do so through Model.steps, similar to sklearn.pipeline.Pipeline.steps. For instance, the code below shows how to retrieve the estimator.


   best_run, model = parent_run.get_output()
   estimator = model.steps[-1]

get_run_sdk_dependencies

Get the SDK run dependencies for a given run.

get_run_sdk_dependencies(iteration=None, check_versions=True, **kwargs)

Parameters

Name Description
iteration
int

The iteration number of the fitted run to be retrieved. If None, retrieve the parent environment.

Default value: None
check_versions
bool

If True, check the versions with current environment. If False, pass.

Default value: True

Returns

Type Description
dict

The dictionary of dependencies retrieved from RunHistory.

Exceptions

Type Description
ConfigException

pause

Return True if the AutoML run was paused successfully.

This method is not implemented.

pause()

Exceptions

Type Description
NotImplementedError

register_model

Register the model with AzureML ACI service.

register_model(model_name=None, description=None, tags=None, iteration=None, metric=None)

Parameters

Name Description
model_name
str

The name of the model being deployed.

Default value: None
description
str

The description for the model being deployed.

Default value: None
tags
dict

Tags for the model being deployed.

Default value: None
iteration
int

Override for which model to deploy. Deploys the model for a given iteration.

Default value: None
metric
str

Override for which model to deploy. Deploys the best model for a different metric.

Default value: None

Returns

Type Description
<xref:Model>

The registered model object.

resume

Return True if the AutoML run was resumed successfully.

This method is not implemented.

resume()

Exceptions

Type Description
NotImplementedError:

retry

Return True if the AutoML run was retried successfully.

This method is not implemented.

retry()

Exceptions

Type Description
NotImplementedError

summary

Get a table containing a summary of algorithms attempted and their scores.

summary()

Returns

Type Description
DataFrame

Pandas DataFrame containing AutoML model statistics.

wait_for_completion

Wait for the completion of this run.

Returns the status object after the wait.

wait_for_completion(show_output=False, wait_post_processing=False)

Parameters

Name Description
show_output
bool

Indicates whether to show the run output on sys.stdout.

Default value: False
wait_post_processing
bool

Indicates whether to wait for the post processing to complete after the run completes.

Default value: False

Returns

Type Description
dict

The status object.

Exceptions

Type Description
ExperimentExecutionException

Attributes

run_id

Return the run ID of the current run.

Returns

Type Description
str

The run ID of the current run.