Using Estimator or SciptRunConfig for Pipeline with Hyperdrive and XGBoost?

Question

I'm learning Azure ML and I'm trying to make a pipeline with HyperDrive and an Xgboost estimator. But I cannot figure out how I add the custom XGBoost environment to my HyperDriveConfig.

Estimator is apparently deprecated and telling me to use ScriptRunConfig instead.

So I have created an Enviroment pointing to a yaml file with dependencies and a ScriptRunConfig pointing to the environment. But how should I use the ScriptRunConfig in HyperDriveConfig?
This is the code I'm trying atm:

env = Environment.from_conda_specification("xgboost", "environment.yml")

src = ScriptRunConfig(
                    source_directory='.',
                    script='train.py',
                    compute_target=compute_target,
                    environment=env,
)

hyperdrive_run_config = HyperDriveConfig(
                                         hyperparameter_sampling=ps, 
                                         primary_metric_name='Accuracy',
                                         primary_metric_goal=PrimaryMetricGoal.MAXIMIZE,
                                         max_total_runs=100,
                                         max_concurrent_runs=4,
                                        #run_config = aml_run_config,
                                        policy=None,
                                        estimator=src
                                        )

When I submit the pipeline get the following error:

AttributeError: 'ScriptRunConfig' object has no attribute '_get_script_run_config'

Answer 1

@Michael Søegaard I think the ScriptRunConfig() should be passed to the run_config parameter instead of the estimator. Here is the sample

from azureml.train.hyperdrive import HyperDriveConfig  
hd_config = HyperDriveConfig(run_config=src,  
                             hyperparameter_sampling=param_sampling,  
                             policy=early_termination_policy,  
                             primary_metric_name="accuracy",  
                             primary_metric_goal=PrimaryMetricGoal.MAXIMIZE,  
                             max_total_runs=100,  
                             max_concurrent_runs=4)