You already have the Python scripts for several steps (including data ingestion, data cleansing, dividing data into train and test sets etc.) of your machine learning tasks but you want to combine them into a consistent, repeatable flow. You want to make use of the orchestration services offered by Azure ML pipelines. You have defined 3 three steps, each of them referencing a piece of your Python code: 
<pre class="brush:java;">step1 = PythonScriptStep(name="train_step",
 script_name="train.py",
 compute_target=aml_compute,
 source_directory=source_directory,
 allow_reuse=False)
step2 = PythonScriptStep(name="compare_step",
 script_name="compare.py",
 compute_target=aml_compute_cluster2,
 source_directory=source_directory,
 allow_reuse=False)
step3 = PythonScriptStep(name="extract_step",
 script_name="extract.py",
 compute_target=aml_compute,
 source_directory=source_directory,
 runconfig=run_config)</pre>
Which parts of the codes should be changed to ensure optimal performance?

Answer options:

A.allow_reuse should be set True for the 1st step of a pipeline
B.source_directory should reference to different folders for each step
C.compute _target should be the same compute for each step in a pipeline
D.parameter runconfig should be set for each step

Answer: B
Option A is incorrect because there is no such constraint for allow_reuse. This is an optional parameter with default value of True. It determines whether outputs of the step’s previous run can be used by the following steps in order to save execution time. Either True or False can be correct.
Option B is CORRECT because in order to optimize the behavior of the pipeline during runs, the recommended practice is to use separate folders for storing scripts and its dependent files for each step. These folders should bethe source_directory for the steps. This way, the size of the snapshot created for the step can be reduced.
Option C is incorrect because there is a compute target at pipeline level which is used by the steps unless specified otherwise at step level. Steps with individual compute requirements can define their own compute target.
Option D is incorrect because the runconfig parameter can be used to specify additional requirements for the run, such as conda dependencies (e.g. ‘scikit-learn’). When missing, a default runconfig will be created. See step parameters below:
PythonScriptStep(script_name, name=None, arguments=None, compute_target=None, runconfig=None, runconfig_pipeline_params=None, inputs=None, outputs=None,params=None, source_directory=None, allow_reuse=True, version=None, hash_paths=None) 
Reference:

https://docs.microsoft.com/en-us/azure/machine-learning/how-to-create-your-first-pipeline
https://github.com/MicrosoftLearning/DP100/blob/master/06A%20-%20Creating%20a%20Pipeline.ipynb
https://github.com/MicrosoftLearning/DP100/blob/master/labdocs/Lab06A.md