Merge pull request #232382 from Blackmist/65566-known-issue

documenting v1 known issue

Author: v-regandowner

articles/machine-learning/v1/how-to-debug-pipelines.md

@@ -128,7 +128,7 @@ file_path = os.path.join(script_dir, "<file_name>")
 - `run_invocation_timeout`: The `run()` method invocation timeout in seconds. (optional; default value is `60`)
 - `run_max_try`: Maximum try count of `run()` for a mini-batch. A `run()` is failed if an exception is thrown, or nothing is returned when `run_invocation_timeout` is reached (optional; default value is `3`). 
 
-You can specify `mini_batch_size`, `node_count`, `process_count_per_node`, `logging_level`, `run_invocation_timeout`, and `run_max_try` as `PipelineParameter`, so that when you resubmit a pipeline run, you can fine-tune the parameter values. In this example, you use `PipelineParameter` for `mini_batch_size` and `Process_count_per_node` and you will change these values when resubmit a run later. 
+You can specify `mini_batch_size`, `node_count`, `process_count_per_node`, `logging_level`, `run_invocation_timeout`, and `run_max_try` as `PipelineParameter`, so that when you resubmit a pipeline run, you can fine-tune the parameter values. In this example, you use `PipelineParameter` for `mini_batch_size` and `Process_count_per_node` and you will change these values when you resubmit a run later. 
 
 ### Parameters for creating the ParallelRunStep
 
@@ -267,6 +267,86 @@ For more information on using the OpenCensus Python library in this manner, see
 
 In some cases, you may need to interactively debug the Python code used in your ML pipeline. By using Visual Studio Code (VS Code) and debugpy, you can attach to the code as it runs in the training environment. For more information, visit the [interactive debugging in VS Code guide](../how-to-debug-visual-studio-code.md#debug-and-troubleshoot-machine-learning-pipelines).
 
+## HyperdriveStep and AutoMLStep fail with network isolation
+
+After using HyperdriveStep and AutoMLStep, when you attempt to register the model you may receive an error.
+
+* You are using Azure Machine Learning SDK v1.
+* Your Azure Machine Learning workspace is configured for network isolation (VNet).
+* Your pipeline attempts to register the model generated by the previous step. For example, in the following code the `inputs` parameter is the saved_model from a HyperdriveStep:
+
+    ```python
+    register_model_step = PythonScriptStep(script_name='register_model.py',
+                                       name="register_model_step01",
+                                       inputs=[saved_model],
+                                       compute_target=cpu_cluster,
+                                       arguments=["--saved-model", saved_model],
+                                       allow_reuse=True,
+                                       runconfig=rcfg)
+    ```
+
+### Workaround
+
+> [!IMPORTANT]
+> This behavior does not occur when using Azure Machine Learning SDK v2.
+
+To work around this error, use the [Run](/python/api/azureml-core/azureml.core.run.run) class to get the model created from the HyperdriveStep or AutoMLStep. The following is an example script that gets the output model from a HyperdriveStep:
+
+```python
+%%writefile $script_folder/model_download9.py
+import argparse
+from azureml.core import Run
+from azureml.pipeline.core import PipelineRun
+from azureml.core.experiment import Experiment
+from azureml.train.hyperdrive import HyperDriveRun
+from azureml.pipeline.steps import HyperDriveStepRun
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        '--hd_step_name', 
+        type=str, dest='hd_step_name', 
+        help='The name of the step that runs AutoML training within this pipeline')
+        
+        
+    
+    args = parser.parse_args()
+    
+    current_run = Run.get_context()
+
+    pipeline_run = PipelineRun(current_run.experiment, current_run.experiment.name)
+
+    hd_step_run = HyperDriveStepRun((pipeline_run.find_step_run(args.hd_step_name))[0])
+    hd_best_run = hd_step_run.get_best_run_by_primary_metric()
+
+    print(hd_best_run)
+    hd_best_run.download_file("outputs/model/saved_model.pb", "saved_model.pb")
+    
+    
+    print("Successfully downloaded model") 
+```
+
+The file can then be used from a PythonScriptStep:
+
+```python
+from azureml.pipeline.steps import PythonScriptStep
+conda_dep = CondaDependencies()
+conda_dep.add_pip_package("azureml-sdk")
+conda_dep.add_pip_package("azureml-pipeline")
+
+rcfg = RunConfiguration(conda_dependencies=conda_dep)
+
+model_download_step = PythonScriptStep(
+    name="Download Model 9",
+    script_name="model_download9.py", 
+    arguments=["--hd_step_name", hd_step_name],
+    compute_target=compute_target,
+    source_directory=script_folder,
+    allow_reuse=False,
+    runconfig=rcfg
+)
+```
+
 ## Next steps
 
 * For a complete tutorial using `ParallelRunStep`, see [Tutorial: Build an Azure Machine Learning pipeline for batch scoring](../tutorial-pipeline-batch-scoring-classification.md).
@@ -275,4 +355,4 @@ In some cases, you may need to interactively debug the Python code used in your
 
 * See the SDK reference for help with the [azureml-pipelines-core](/python/api/azureml-pipeline-core/) package and the [azureml-pipelines-steps](/python/api/azureml-pipeline-steps/) package.
 
-* See the list of [designer exceptions and error codes](../algorithm-module-reference/designer-error-codes.md).
+* See the list of [designer exceptions and error codes](../algorithm-module-reference/designer-error-codes.md).

articles/machine-learning/v1/how-to-use-automlstep-in-pipelines.md

@@ -366,6 +366,10 @@ print("Registered version {0} of model {1}".format(model.version, model.name))
 
 ### Write the PythonScriptStep code
 
+
+> [!WARNING]
+> If you are using the Azure Machine Learning SDK v1, and your workspace is configured for network isolation (VNet), you may receive an error when running this step. For more information, see [HyperdriveStep and AutoMLStep fail with network isolation](how-to-debug-pipelines.md#hyperdrivestep-and-automlstep-fail-with-network-isolation).
+
 The model-registering `PythonScriptStep` uses a `PipelineParameter` for one of its arguments. Pipeline parameters are arguments to pipelines that can be easily set at run-submission time. Once declared, they're passed as normal arguments. 
 
 ```python