TypeError related to workspace while building pipeline - Python SDK

Question

Hi! I got this error while trying to run a job for creating a pipeline using Python SDK:

Traceback (most recent call last):

  Cell In[14], line 3
    new_pipeline = Pipeline(workspace=ws, steps=steps)

  File ~\anaconda3\Lib\site-packages\azureml\core\_experiment_method.py:104 in wrapper
    return init_func(self, *args, **kwargs)

  File ~\anaconda3\Lib\site-packages\azureml\pipeline\core\pipeline.py:95 in __init__
    self._graph_context = _GraphContext("placeholder", workspace=workspace,

  File ~\anaconda3\Lib\site-packages\azureml\pipeline\core\_graph_context.py:44 in __init__
    self._workflow_provider = _AevaWorkflowProvider.create_provider(workspace=workspace,

  File ~\anaconda3\Lib\site-packages\azureml\pipeline\core\_aeva_provider.py:104 in create_provider
    service_caller = _AevaWorkflowProvider.create_service_caller(workspace=workspace,

  File ~\anaconda3\Lib\site-packages\azureml\pipeline\core\_aeva_provider.py:121 in create_service_caller
    service_endpoint = _AevaWorkflowProvider.get_endpoint_url(workspace, experiment_name)

  File ~\anaconda3\Lib\site-packages\azureml\pipeline\core\_aeva_provider.py:138 in get_endpoint_url
    service_context = ServiceContext(

TypeError: ServiceContext.__init__() missing 1 required positional argument: 'workspace_discovery_url'

The error seems to be related with my workspace but I really don't see the problem. Here is my code for the pipeline creation and my .config file for worspace creation.

#------------------------------------------------------------
# Job script build a pipeline and running it in a experiment
#------------------------------------------------------------

from azureml.core import Workspace

ws = Workspace.from_config("./config")

#------------------------------------------------------------

# Create custom environment

from azureml.core import Environment
from azureml.core.conda_dependencies import CondaDependencies

myenv = Environment(name="MyEnvironment")

#Create the dependencies object
myenv_dep = CondaDependencies.create(conda_packages=["scikit-learn", "pandas", "numpy", "scipy"])

myenv.python.conda_dependencies = myenv_dep

# Register the environment
myenv.register(ws)

#------------------------------------------------------------

# Create compute cluster

# Provisioning configuration using AMLCompute
from azureml.core.compute import AmlCompute

cluster_name = "pipeline-cluster"

compute_config = AmlCompute.provisioning_configuration(
                            vm_size="STANDARD_E2d_v4",
                            max_nodes=2)


from azureml.core.compute import ComputeTarget
compute_cluster = ComputeTarget.create(ws, cluster_name, compute_config)

compute_cluster.wait_for_completion()


#------------------------------------------------------------
# Configuration for the steps

from azureml.core.runconfig import RunConfiguration
run_config = RunConfiguration()
run_config.target = compute_cluster
run_config.environment = myenv


#------------------------------------------------------------
# Defining pipeline steps

from azureml.pipeline.steps import PythonScriptStep
from azureml.pipeline.core import PipelineData
from azureml.data.data_reference import DataReference

#input_ds = ws.datasets.get("Defaults")

#Create input as DataReference for using in step 1
input_data = DataReference(datastore=ws.get_default_datastore(),
                                data_reference_name="Defaults",
                                path_on_datastore="defaults.csv")


dataFolder = PipelineData("dataFolder", datastore=ws.get_default_datastore())


# Step 1 - DATA PREPARATION

dataPrep_step = PythonScriptStep(name="01 Data Preparation",
                                 source_directory=".",
                                 script_name="13-auto-dataprep",
                                 inputs=[input_data],
                                 outputs=[dataFolder],
                                 runconfig=run_config,
                                 arguments=["--dataFolder", dataFolder]
                                 )

# Step 2 - TRAIN THE MODEL

train_step = PythonScriptStep(name="02 Train the Model",
                                 source_directory=".",
                                 script_name="14-auto-training",
                                 inputs=[dataFolder],
                                 runconfig=run_config,
                                 arguments=["--dataFolder", dataFolder]
                                 )

#------------------------------------------------------------

# Configure and build the pipeline
steps = [dataPrep_step, train_step]
from azureml.pipeline.core import Pipeline
new_pipeline = Pipeline(workspace=ws, steps=steps)


# Create the experiment and run the pipeline
from azureml.core import Experiment
new_experiment = Experiment(workspace=ws, name="PipelineExp01")
new_pipeline_run = new_experiment.submit(new_pipeline)

new_pipeline_run.wait_for_completion(show_output=True)

Hope someone can help me!

Answer 1

Valeria Gonzalez Guerrero Thanks for the details, Are you using the SDK v1 or v2? If you are using v2, You can install the Install the azure ml SDK v2 using the following command pip install azure-ai-ml.

Here is the document for the installation. https://learn.microsoft.com/en-us/python/api/overview/azure/ai-ml-readme?view=azure-python

If you are using the v1 SDK for the automl is outdated and not supported now.

We are able to run the pipeline/other samples and connect to workspace as shown below.