Lab400.md

ML Models in Synapse TODO

Quick Overview:

Using Azure Synapse Analytics, data scientists are no longer required to use separate tooling to create and deploy machine learning models.

In this exercise, you will create multiple machine learning models. You will learn how to consume these models in your notebook. You will also deploy a model as a web service to Azure Container Instances and consume the service.

Task 1: Create a SQL Datastore and source Dataset

1. Open the lab resource group, locate and open the amlworkspace{{suffix}} Machine Learning resource.

   

2. On the Overview screen of the Machine Learning resource, select the Studio web URL link.

   

3. From the left menu of Azure Machine Learning Studio, select the Datastores item.

   

4. On the Datastores screen top menu, select + New datastore.

5. On the New datastore blade, configure it as follows and select Create:

   Field	Value
   New datastore (name)	sqlpool01
   Datastore type	Azure SQL database
   Account selection method	From Azure subscription
   Subscription ID	Select the lab subscription.
   Server name / database name	Select asaworkspace{{suffix}}/SQLPool01.
   Authentication type	SQL authentication
   User ID	asa.sql.admin
   Password	The SQL Admin password you chose when deploying the lab resources.

   

6. From the left menu, select Datasets, and with the Registered datasets tab selected, expand the + Create dataset button and select From datastore.

   

7. In the Create dataset from datastore Basic info form, name the dataset AggregatedProductSeasonality and select Next.

   

8. On the Datastore selection form, select Previously created datasource, choose sqlpool01 from the list and select the Select datasource button.

   

9. In the next Datastore selection form, enter the following SQL query. Then expand the Advanced settings and enter 100 for the Query timeout (seconds) value. Select Next:

   SELECT  P.ProductId,P.Seasonality,S.TransactionDateId,COUNT(*) as TransactionItemsCount
   FROM wwi_mcw.SaleSmall S
   JOIN wwi_mcw.Product P ON S.ProductId = P.ProductId
   GROUP BY P.ProductId ,P.Seasonality,S.TransactionDateId

   

10. The Settings and preview data table will be displayed after a few moments. Review this data, then select the Next button.

    

11. Review the Schema field listing, then select Next.

    

12. On the Confirm details screen, select Create.

    

Task 2: Create compute infrastructure

1. From the left menu of Machine Learning Studio, select Compute.

2. On the Compute screen with the Compute instances tab selected. Choose the Create button.

   

3. On the Create compute instance, Select virtual machine form, configure it as follows, then select Next:

   Field	Value
   Virtual machine type	CPU
   Virtual machine size	Search for and select Standard_DS3_v2.

   

4. On the Configure Settings form, enter a globally unique Compute name of your choice, and select Create.

   

5. Select the Compute clusters tab, and select Create.

6. On the New compute cluster, Select virtual machine form, configure the virtual machine as follows, then select Next:

   Field	Value
   Virtual machine priority	Dedicated
   Virtual machine type	CPU
   Virtual machine size	Search for and select Standard_DS3_v2.

   

7. On the Configure Settings form, configure it as follows, then select Create:

   Field	Value
   Compute name	automlcluster
   Minimum number of nodes	0
   Maximum number of nodes	3
   Idle seconds before scale down	120

   

Task 3: Use a notebook in AML Studio to prepare data and create a Product Seasonality Classifier model using XGBoost

1. In Azure Machine Learning (AML) Studio, select Notebooks from the left menu.

2. In the Notebooks pane, select the Upload icon from the toolbar.

   

3. In the Open dialog, select Hands-on lab/artifacts/ProductSeasonality_sklearn.ipynb. When prompted, check the boxes to Overwrite if already exists and I trust contents of this file and select Upload.

   

4. In the top toolbar of the notebook, expand the Editors item, and select Edit in Jupyter.

   

5. Review and run each cell in the notebook individually to gain understanding of the functionality being demonstrated.

Note: Running this notebook in its entirety is required for the next task.

Task 4: Leverage Automated ML to create and deploy a Product Seasonality Classifier model

1. In Azure Machine Learning (AML) Studio, select Experiments from the left menu, then expand the + Create button, and select Automated ML run.

   

2. In the previous task, we registered our PCA dataframe (named pcadata) to use with Auto ML. Select pcadata from the list and select Next.

   

3. On the Configure run screen, select the Create a new compute link beneath the Select compute cluster field.

4. Back on the Configure run form, name the experiment ProductSeasonalityClassifier, select Seasonality as the Target column and select automlcluster as the compute cluster. Select Next.

   

5. On the Select task type screen, select Classification, then choose Finish.

   

6. The experiment will then be run. It will take approximately 20-25 minutes for it to complete. Once it has completed, it will display the run results details. In the Best model summary box, select the Algorithm name link.

   

7. On the Model run screen, select Deploy from the top toolbar.

   

8. On the Deploy a model blade, configure the deployment as follows, then select Deploy:

   Field	Value
   Name	productseasonalityclassifier
   Description	Product Seasonality Classifier.
   Compute type	Azure Container Instance
   Enable authentication	Off

   

9. Once deployed, the Model summary will be displayed. You can view the endpoint by selecting the Deploy status link.

   

10. Review the details of the deployed model service endpoint.