Objective
The objective of this demo asset is to demonstrate building a predictive model with Spark machine learning API (SparkML) to predict customer churn, and deploy it for scoring in Machine Learning (ML).
This repository contains the following assets for the Telco Churn demo in DSX Local
Demo setup
-
Create a DSX project and name it "DSX Local Lab - Telco Churn"
-
Import
Tip: First download the csv files before importing them into your project. When downloading the csv files, make sure to click the Raw button to display the data in its raw format, right-click and select "Save Page As". -
Import notebook
Within the "DSX Local Lab - Telco Churn" project, add a Notebook and choose to import it from this URL: https://github.com/elenalowery/DSX-Local-Telco-Churn/blob/master/Notebooks/Telco%20Churn%20ML_Local.ipynb -
Follow instructions in the notebook to add project token, and work through the notebook
-
Optional: deploy Telco Churn UI application: instructions in the folder. Or you can use this deployed UI: https://predictcustomerchurn.mybluemix.net/
-
If you would like to show Model Management - create several deployments from this or different notebooks
-
Optionally, watch a video of the presentation and demo
Demo
- Follow the agenda in the presentation
- During the demo show capabilities of DSX in the context of Telco Churn use case
- Start with the overview of the use case and optionally the Telco Churn UI
- Log in to DSX Local and create a new project
- Show collaboration features for the project
- Load data and explain what type of data sources are supported
- Create a notebook from File
- Walk through the notebook
- Explain the deployment process - via UI and API
- Test the model via UI or API and explian how the demo UI makes the same call
- Wrap up with architecture discussion
Converting the notebook to use HDFS data sources
- Load .csv files into HDFS
- Make a copy of the notebook or use the sample notebook in the Notebooks folder
- Replace Object Storage access code with HDFS access code