Community Articles

Find and share helpful community-sourced technical articles.
Announcements
Celebrating as our community reaches 100,000 members! Thank you!
avatar
Cloudera Employee

Table of Contents

Use Case

As a healthcare provider / public health official, I want to respond equitably to the COVID-19 pandemic as quickly as possible, and serve all the communities that are adversely impacted in the state of California.
I want to use health equity data reported by California Department of Public Health (CDPH) to identify impacted members and accelerate the launch of outreach programs.

Design

Collect - Ingest data from https://data.chhs.ca.gov/dataset/covid-19-equity-metrics using NiFi.
Enrich - Transform the dataset using Spark and load Hive tables.
Report - Gather insights using Hive tables and Data Visualization.
Predict - Connect to Hive tables and build Machine Learning (ML) models of your choice.

 

 

Implementation

Prerequisites

 

Steps to create this data pipeline, are as follows:

Please note that this data pipeline's documentation is in accordance with CDP Runtime Version 7.2.12.

Step #1 - Setup NiFi Flow

  • Create or use a Data Hub Cluster with NiFi.
    Following Data Hub Cluster type can be used for this exercise - "7.2.12 - Flow Management Light Duty with Apache NiFi, Apache NiFi Registry".

  • Go to NiFi user interface and upload NiFi-CDPH.xml as a template.

  • NiFi-CDPH.xml uses PutS3Object processor to connect to an existing Amazon S3 bucket. Please change the properties in this processor to use your own bucket.

  • If you don't use Amazon S3 storage, please replace PutS3Object processor with a processor of your own choice. Refer NiFi docs for details.
    For quick reference, here are the frequently used processors to write to a file system -

    • PutAzureDataLakeStorage for Azure Data Lake Storage
    • PutGCSObject for Google Cloud Storage
    • PutHDFS for Hadoop Distributed File System (HDFS)
  • Execute the flow and ensure InvokeHTTP processors are able to get covid19case_rate_by_social_det.csv and covid19demographicratecumulative.csv. Verify that these files are added to your storage bucket.

  • Once you're satisfied with functions of this NiFi flow, download the flow definition.

  • For reference, here's a picture of the flow in NiFi user interface -


 

Step #2 - Setup Cloudera DataFlow (CDF)

  • Now that NiFi flow is ready, it's time to deploy it in your CDP environment. Go to CDF user interface, and ensure CDF service is enabled in your CDP environment.
  • Import flow definition.
  • Select imported flow, click on Deploy and follow the wizard to complete the deployment. Please note that Extra Small NiFi node size is enough for this data ingestion.
  • After deployment is done, you would see the flow in Dashboard. You will be able to manage deployment of your flow in the Dashboard and perform functions like start/terminate flow, view flow, change runtime, view KPIs, view Alerts, etc.
  • In Step #1, you've already executed the NiFi flow to add the source files to your storage bucket. So, you don't need to execute it again from CDF. But even if you do, it's going to just overwrite the files and not hurt anything.

Step #3 - Setup Cloudera Data Engineering (CDE)

  • Go to CDE user interface, and ensure CDE service is enabled in your CDP environment & a virtual cluster is available for use.
  • Create a Spark job. In the wizard, upload enrich.py program and leave other options as default. Please change the fs variable in enrich.py program to point to your bucket.
  • Execute the job and monitor logs to ensure it's finished successfully. It takes approx. 4 minutes to finish.
  • Following Hive tables are created by this job:
    • cdph.data_dictionary
    • cdph.covid_rate_by_soc_det
    • cdph.covid_demo_rate_cumulative
    • member.member_profile
    • member.target_mbrs_by_income
    • member.target_mbrs_by_age_group

Step #4 - Setup Cloudera Data Warehouse (CDW)

  • Go to CDW user interface. Ensure CDW service is activated in your CDP environment, and a Database Catalog & a Virtual Warehouse compute cluster are available for use.
  • Open Hue editor and explore the Hive tables created by CDE job.
    -- Raw Data
    select * from cdph.data_dictionary a;
    select * from cdph.covid_rate_by_soc_det a;
    select * from cdph.covid_demo_rate_cumulative a;
    select * from member.member_profile a;

Step #5 - Setup Cloudera Data Visualization (Data VIZ) Dashboard

  • Go to Data VIZ user interface.

  • Under the DATA tab, create first Dataset - COVID Rate by Social Determinants

Dataset Details:

 

 

Update Dimensions & Measures to look like below:

 

 

  • Under the DATA tab, create second Dataset - COVID Demographic Rate Cumulative

Dataset Details:

 

 

Update Dimensions & Measures to look like below:

 

 

  • Once Datasets are available, go to VISUALS tab and create a new dashboard.

  • Let's create first visual in the dashboard, to show COVID-19 cases by income-groups. Select Default Hive VW and COVID Rate by Social Determinants from drop down menus, and create a new visual. Set the following parameters -

    • Visual Type - Combo (Combined Bar/Line)
    • Dimension - priority_sort
    • Bar Measure - avg(case_rate_per_100k)
    • Tooltips - max(social_tier)
    • Filters - social_det in ('income')

 

 

  • Let's create second visual in the dashboard, to show COVID-19 related deaths by age-groups. Select Default Hive VW and COVID Demographic Rate Cumulative from drop down menus, and create a new visual. Set the following parameters -

    • Visual Type - Lines
    • X Axis - demographic_set_category. Go to Field Properties, and select "Ascending" under "Order and Top K".
    • Y Axis - avg(metric_value_per_100k)
    • Filters -
      • demographic_set in ('age_gp4')
      • metric in ('deaths')
      • county in ('Alameda', 'Contra Costa', 'Los Angeles', 'San Diego', 'San Francisco', 'Ventura'). To see data for all counties in California, USA, remove this filter.

 

  • For reference, here's the complete dashboard:

 

 

Step #6 - Identify impacted members in Hue editor

  • As you can see in the visuals, below $40K is the most impacted income group in terms of COVID-19 cases, and 65+ is the most impacted age group in terms of COVID-19 related deaths. You can now use this information, to filter members that are in these categories.
  • Open Hue editor and execute the following queries to get impacted members:
    select * from member.target_mbrs_by_income a where social_tier = 'below $40K';
    select * from member.target_mbrs_by_age_group a where demographic_set_category = '65+';

Step #7 - View Hive tables in Cloudera Data Catalog

  • Go to Data Catalog user interface. Select any Hive table created in this exercise and see its lineage, schema, audits, etc.

Step #8 - Setup Cloudera Machine Learning (CML)

  • Go to CML user interface. Under ML Workspaces menu item, provision a new workspace. While provisioning a new workspace, enable Advanced Options and check "Enable Public IP Address for Load Balancer". This could take ~45 minutes to finish.

  • Once workspace is available, create a New Project. Under Initial Setup, Template tab is selected by default, that works for most users. But you also have options to start from scratch (Blank), use existing Applied Machine Leaning Prototypes (AMPs - see AMPs navigation menu item for details), use local files (Local Files) or Git repository (Git).

  • Download covid_outreach.ipynb and upload it in your project.

  • If multiple people are going to work on this project, add them as collaborators with the right role under Collaborators menu item.

  • Once you have the project setup, start a New Session. Select JupyterLab in Editor dropdown and check Enable Spark.

 

  • In your session, select covid_outreach.ipynb notebook. Please replace YOUR_USERNAME and YOUR_PASSWORD in the notebook with your workload's credentials. If you're not sure how to setup this up, refer Setting the workload password.

 

  • Execute the notebook and see data in Hive tables.

  • Now, you're ready to play around with the datasets and build your ML models.

  • Run Experiment under Experiments menu item when you have a draft model ready.

  • When you're ready to deploy the model, go to Models menu item and select New Model.

  • Once you're satisfied with the results of your model, create a New Job under Jobs menu item to setup arguments, schedule, notifications & so on.

  • For reference, please see menu items highlighted in Blue box that are referred in prior bullet points.

 

573 Views
0 Kudos