Webinar - Enrich Tableau with ML - Schindlauer (20160615)

18
Tableau & Dato Predictive Services Machine Learning in Tableau with Python

Transcript of Webinar - Enrich Tableau with ML - Schindlauer (20160615)

Tableau &Dato Predictive ServicesMachine Learning in Tableau with Python

2

Hello my name is..

Roman SchindlauerSenior Product Manager, Dato

3

Hello my name is..

Bora BeranStaff Product Manager, Tableau

4

Agenda• Overview• Demos:

- Clustering- Sentiment Analysis- Lead Scoring

• Best Practices

5

Dato & Tableau• The power of Python in Tableau dashboards• Tableau 8.1: R• Tableau 10: Python

• Scenarios:- Run complex models on user interaction data (churn prediction,

lead scoring, sentiment analysis, risk analysis)- Process workbook data through an expressive programming

language- Share external functions and models

Clustering with scikit-learn

7

Dato Predictive Services• Machine Learning in production

Load Balancer

ApplicationPython codeModel queries

Model ServerModel Server

Metrics Logs

Cache

Python execution environment

Model server

Managementmodels, ops

8

Predictive Service

Tableau Integration• Execute code in a PS

Tableau

SCRIPT( ...)

Python execution environment

ML modelML modelML model

Sentiment Analysis with GraphLab Create

10

PS as a Model Server• Persist code in the server, call like a function

• Discover, share and re-use• Update without affecting clients

Predictive ServiceTableau

SCRIPT( ...)

Python execution environment

ML modelML modelML model

Deploying a trained model

12

Discovering deployed models• Workbook Dato Model Catalog:

- Uses a WDC that talks to a PS- Lists deployed models- Ideally, model author has added metadata to the

service- Simplifies the SCRIPT authoring

13

Service Management & Monitoring• Connect:

ps = gl.deploy.predictive_service.load(“s3://…”)

• Status:ps.get_status()

• Scale the service:ps.add_nodes(1)ps.set_scale_factor(4)

• Get metrics:ps.get_metrics(name="test", start_time=“6h”)

Need path and access to state file!

14

Summary• ML integration

- Execute any Python code- Call deployed models (Discover through WDC-based catalog)- Data type compatibility

• Efficiency- Define correct table calculations- Scale the predictive service- Monitor PS health

• Security- Querying: API key is auto-generated, can be changed- Administration: PS admins need access to state path- SSL is supported

15

Further sources• http://www.tableau.com/support/desktop• https://dato.com/learn/

- Notebook gallery- User guide- API reference

• Dato Webinars

● 2 days with 60 talks and hands-on tutorials● Tools, techniques, latest research & best practices● Speakers from Pinterest, Pandora, Kaggle, Google,

UW, Quora, Salesforce, Uber, Tableau, Stanford, CMU and more

Data Science SummitRegister Now

http://bit.ly/DSS-SF-2016

Questions?

Thank you for joining us!