Assigning Version to Custom Environment in Azure.ai.ml Command Execution

Question

Assigning Version to Custom Environment in Azure.ai.ml Command Execution

Saeed Misaghian 20

I am currently encountering an issue that I have been unable to resolve. Despite seeming straightforward, it has proven to be quite challenging.

I am attempting to execute a Python script via a command from azure.ai.ml. My script incorporates an XGBoost model, which necessitates a customized environment. I have made several attempts to customize the environment within the command, but each attempt has resulted in an error message indicating that I need to assign a version to the environment. Interestingly, the Environment class does not accept a version as it is not an argument within that class.

Could you please provide guidance on how I can resolve this issue?

I appreciate your assistance and look forward to your response.

Best regards, Saeed

custom_env = Environment.from_conda_specification(name="custom_env", file_path="conda_env.yml")
custom_env.register(workspace=ws)

# configure job

job = command(
    code="./src",
    command="python train-model-script.py --training_data ./data/energy_data.xlsx",
    environment="env-name",
    compute="aml-cluster",
    display_name="building-energy-train",
    experiment_name="building-energy-training", 
    tags={"model_type": "XGBOOST", "model":"Building energy dataset"}
    )

# submit job
returned_job = ml_client.create_or_update(job)
aml_url = returned_job.studio_url


yamle file: 
name: my_env
dependencies:
  - python
  - scikit-learn
  - xgboost
  - argparse
  - numpy
  - pandas
  - matplotlib

dupammi 8,615 Reputation points Microsoft External Staff

2023-12-14T07:21:01.0833333+00:00
Hi @Saeed Misaghian ,

Thank you for using the Microsoft Q&A.

To address your query regarding the version, create a new environment named "myenv" from a Conda specification file, as shown below:

from azureml.core import Workspace, Experiment, Environment from azureml.core import ScriptRunConfig, Experiment # Specify the full path to the config.json file config_path = "/home/azureuser/cloudfiles/code/Users/v-dupammi/outputs/" # Load the workspace from configuration ws = Workspace.from_config(path=config_path) custom_env = Environment.from_conda_specification(name="myenv", file_path="myenv.yml") custom_env.register(workspace=ws)

myenv.yml

name: my_env dependencies: - python=3.6.2 - azureml-defaults~=1.34.0 - azureml-monitoring - torch - pytorch-pretrained-bert==0.4.0

Once registered, then use the new environment in the job by specifying the environment name in the environment parameter of the command function.

Please refer this Azure official documentation for more details on using the environment in the command.

I hope this helps! Thank you!
dupammi 8,615 Reputation points Microsoft External Staff

2023-12-15T10:13:58.2433333+00:00

Hi @Saeed Misaghian,

We haven’t heard from you on the last response and was just checking back to see if you got a chance to check my response above.

Thank you!
Saeed Misaghian 20 Reputation points

2023-12-15T10:18:08.1833333+00:00

Hello @dupammi

Thank you very much for the response and the follow up. Can you please give me a couple of hours to check in the response? Unfortunately I didn't have time after your comment to check in the process. I will update you today for sure!

Thanks a million

Saeed

Saeed Misaghian 20

Hello @dupammi

Sorry for the late reply. I went through what you have mentioned, however the issue still exists.

It says : At least one required parameter is missing\u001b[39m\u001b[39m\n\nDetails: \n\n\u001b[31m(x) Failed to extract version when parsing asset my_env of type environments as arm id. Version must be provided

However, register does not have the version argument at all..

So, I can register the environment but I cannot use it in command.

Looking forward to your opinion on it.

Regards

Saeed

ws = Workspace.get(name=workspace_name,
                          subscription_id=subscription_id,
                          resource_group=resource_group)

# get dataset id 
ml_client = MLClient(
    DefaultAzureCredential(), subscription_id, resource_group, workspace_name
)

data_asset = ml_client.data.get("my-dataset", version="1")

custom_env = Environment.from_conda_specification(name="my_env", file_path="conda_env.yml")
custom_env.register(workspace=ws)

job = command(
    code="./src",
    command="python train-model-mlflow.py --training_data ${{inputs.training_data}}",
    environment="my_env",
    compute="aml-cluster",
    display_name="building-energy-train",
    experiment_name="building-energy-training", 
    tags={"model_type": "xgboost", "model": "my dataset"},
    inputs={'training_data': Input(path=data_asset.id,
              type=AssetTypes.URI_FILE,
              mode=InputOutputModes.RO_MOUNT)}  # Reference the dataset by name
)

# Submit job
returned_job = ml_client.create_or_update(job)
aml_url = returned_job.studio_url
print("Monitor your job at", aml_url)

dupammi 8,615 Microsoft External Staff

Hi @Saeed Misaghian ,

Please try the job submission through ScriptRunConfig.

For this, I created my compute target as shown below:

Compute target creation

from azureml.core.compute import ComputeTarget, AmlCompute
from azureml.core.compute_target import ComputeTargetException

# Choose a name for your CPU cluster
cluster_name = "cluster2"

# Verify that cluster does not exist already
try:
    cluster = ComputeTarget(workspace=ws, name=cluster_name)
    print('Found existing cluster, use it.')
except ComputeTargetException:
    compute_config = AmlCompute.provisioning_configuration(vm_size='STANDARD_D3_V2',
                                                           min_nodes=1,
                                                           max_nodes=4)
    cluster = ComputeTarget.create(ws, cluster_name, compute_config)

cluster.wait_for_completion(show_output=True)

Output

User's image

Submit Job using ScriptRunConfig

# 'myenv' is the name of your environment
myenv = Environment.get(workspace=ws, name='myenv')
# Configure and submit your training run
config = ScriptRunConfig(source_directory='.',
                        script='train.py',  # 'train.py' is training script
                        arguments=['--training_data', 'ner_dataset.csv'],
                        compute_target=cluster,
                        environment=myenv)

# Submit the run to the experiment
experiment = Experiment(ws, 'MyExperiment')
script_run = experiment.submit(config)

Output of Job Submission User's image

Job in portal

User's image

I hope this helps!

Thank you!

Saeed Misaghian 20 Reputation points

2023-12-15T23:17:28.91+00:00

Hello @dupammi

Thank you for your efforts and guidance in resolving this issue.

I attempted to use ScriptRunConfig, but the current workspace I’m using doesn’t permit ScriptRunConfig to create an environment image in the backend. This is because Azure apparently builds a container registry in the backend, which my resource group doesn’t allow. I should create another resource group under my own subscription to test it.

However, would you agree that there’s a bug in the command that prevents passing the environment to it? I’ve flagged this in Microsoft Azure’s Github. Perhaps you could take a look as well!

Regardless, thank you very much. Have a pleasant day or evening, wherever you are! :)

https://github.com/Azure/azure-sdk-for-python/issues/33513

Regards

Saeed

Answer accepted by question author

0 additional answers

Your answer

dupammi 8,615 Reputation points Microsoft External Staff

2023-12-14T07:21:01.0833333+00:00

Hi @Saeed Misaghian ,

Thank you for using the Microsoft Q&A.

To address your query regarding the version, create a new environment named "myenv" from a Conda specification file, as shown below:

from azureml.core import Workspace, Experiment, Environment from azureml.core import ScriptRunConfig, Experiment # Specify the full path to the config.json file config_path = "/home/azureuser/cloudfiles/code/Users/v-dupammi/outputs/" # Load the workspace from configuration ws = Workspace.from_config(path=config_path) custom_env = Environment.from_conda_specification(name="myenv", file_path="myenv.yml") custom_env.register(workspace=ws)

myenv.yml

name: my_env dependencies: - python=3.6.2 - azureml-defaults~=1.34.0 - azureml-monitoring - torch - pytorch-pretrained-bert==0.4.0

Once registered, then use the new environment in the job by specifying the environment name in the environment parameter of the command function.

Please refer this Azure official documentation for more details on using the environment in the command.

I hope this helps! Thank you!
dupammi 8,615 Reputation points Microsoft External Staff

2023-12-15T10:13:58.2433333+00:00

Hi @Saeed Misaghian,

We haven’t heard from you on the last response and was just checking back to see if you got a chance to check my response above.

Thank you!
Saeed Misaghian 20 Reputation points

2023-12-15T10:18:08.1833333+00:00

Hello @dupammi

Thank you very much for the response and the follow up. Can you please give me a couple of hours to check in the response? Unfortunately I didn't have time after your comment to check in the process. I will update you today for sure!

Thanks a million

Saeed
Saeed Misaghian 20 Reputation points

2023-12-15T19:06:46.05+00:00

Hello @dupammi

Sorry for the late reply. I went through what you have mentioned, however the issue still exists.

It says : At least one required parameter is missing\u001b[39m\u001b[39m\n\nDetails: \n\n\u001b[31m(x) Failed to extract version when parsing asset my_env of type environments as arm id. Version must be provided

However, register does not have the version argument at all..

So, I can register the environment but I cannot use it in command.

Looking forward to your opinion on it.

Regards

Saeed

ws = Workspace.get(name=workspace_name, subscription_id=subscription_id, resource_group=resource_group) # get dataset id ml_client = MLClient( DefaultAzureCredential(), subscription_id, resource_group, workspace_name ) data_asset = ml_client.data.get("my-dataset", version="1") custom_env = Environment.from_conda_specification(name="my_env", file_path="conda_env.yml") custom_env.register(workspace=ws) job = command( code="./src", command="python train-model-mlflow.py --training_data ${{inputs.training_data}}", environment="my_env", compute="aml-cluster", display_name="building-energy-train", experiment_name="building-energy-training", tags={"model_type": "xgboost", "model": "my dataset"}, inputs={'training_data': Input(path=data_asset.id, type=AssetTypes.URI_FILE, mode=InputOutputModes.RO_MOUNT)} # Reference the dataset by name ) # Submit job returned_job = ml_client.create_or_update(job) aml_url = returned_job.studio_url print("Monitor your job at", aml_url)
dupammi 8,615 Reputation points Microsoft External Staff

2023-12-15T23:08:14.5233333+00:00

Hi @Saeed Misaghian ,

Please try the job submission through ScriptRunConfig.

For this, I created my compute target as shown below:

Compute target creation

from azureml.core.compute import ComputeTarget, AmlCompute from azureml.core.compute_target import ComputeTargetException # Choose a name for your CPU cluster cluster_name = "cluster2" # Verify that cluster does not exist already try: cluster = ComputeTarget(workspace=ws, name=cluster_name) print('Found existing cluster, use it.') except ComputeTargetException: compute_config = AmlCompute.provisioning_configuration(vm_size='STANDARD_D3_V2', min_nodes=1, max_nodes=4) cluster = ComputeTarget.create(ws, cluster_name, compute_config) cluster.wait_for_completion(show_output=True)

Output

Submit Job using ScriptRunConfig

# 'myenv' is the name of your environment myenv = Environment.get(workspace=ws, name='myenv') # Configure and submit your training run config = ScriptRunConfig(source_directory='.', script='train.py', # 'train.py' is training script arguments=['--training_data', 'ner_dataset.csv'], compute_target=cluster, environment=myenv) # Submit the run to the experiment experiment = Experiment(ws, 'MyExperiment') script_run = experiment.submit(config)

Output of Job Submission

Job in portal

I hope this helps!

Thank you!
Saeed Misaghian 20 Reputation points

2023-12-15T23:17:28.91+00:00

Hello @dupammi

Thank you for your efforts and guidance in resolving this issue.

I attempted to use ScriptRunConfig, but the current workspace I’m using doesn’t permit ScriptRunConfig to create an environment image in the backend. This is because Azure apparently builds a container registry in the backend, which my resource group doesn’t allow. I should create another resource group under my own subscription to test it.

However, would you agree that there’s a bug in the command that prevents passing the environment to it? I’ve flagged this in Microsoft Azure’s Github. Perhaps you could take a look as well!

Regardless, thank you very much. Have a pleasant day or evening, wherever you are! :)

https://github.com/Azure/azure-sdk-for-python/issues/33513

Regards

Saeed

Answer 1

dupammi 8,615 Microsoft External Staff

Hi @Saeed Misaghian ,

You're welcome! I'm pleased to learn that you've successfully identified an issue and have promptly submitted it to Microsoft Azure's GitHub repository.

Thank you for understanding!

If this answers your query, do click Accept Answer and Yes for was this answer helpful.

Deleted

This comment has been deleted due to a violation of our Code of Conduct. The comment was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.

Share via

Assigning Version to Custom Environment in Azure.ai.ml Command Execution

0 additional answers

Your answer