Scribd
Upload
44 views39 pages

Boost Your Analytics With ML For SQL Nerds: SQL Saturday Spokane - Mar 10, 2018

This document discusses how to boost analytics with machine learning for SQL users. It introduces SQL Server 2017 machine learning services with R, which allow users to operationalize machine learning models by running R scripts directly in the database. It covers topics like reducing data movement, deploying models through stored procedures, and achieving enterprise scale. Tips are provided for features like custom SQL Server Management Studio reports and using the SQL Server extensibility framework.

Uploaded by

Manosc La
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
44 views39 pages

Boost Your Analytics With ML For SQL Nerds: SQL Saturday Spokane - Mar 10, 2018

This document discusses how to boost analytics with machine learning for SQL users. It introduces SQL Server 2017 machine learning services with R, which allow users to operationalize machine learning models by running R scripts directly in the database. It covers topics like reducing data movement, deploying models through stored procedures, and achieving enterprise scale. Tips are provided for features like custom SQL Server Management Studio reports and using the SQL Server extensibility framework.

Uploaded by

Manosc La
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 39

Boost your Analytics with ML

for SQL Nerds

SQL Saturday Spokane – Mar 10, 2018


Julie Koesmarno | @MsSQLGirl | mssqlgirl.com | [email protected]
Principal Program Manager in Business Analytics for SQL Products
Business Analytics Lead
Drive business analytics strategy and vision
for Business Growth and Customer Success of
Azure Data products.

Author
SQL Server 2017 Machine Learning Services

Julie Koesmarno with R

Blog: MsSQLGirl.com

Principal Program Manager,


Speaker
Microsoft PASS Summit, SQL Saturdays, User Groups -
Data Warehouse / BI, Data Visualization, SQL
/juliekoesmarno @MsSQLGirl [email protected] Server + R, ADLA
SQL Server 2017 Machine Learning Services with R

e-book $10 (save 68%)


http://bit.ly/2DdfPLs
Agenda

1. Y ML
2. Operationalizing ML
3. Tips & Tricks
4. Resources
1. Y ML?
automation
delighting customers
Intelligent solutions enable differentiation

Deepen Predict Automate


Engagement Outcomes Actions
2. Operationalizing ML
Language Platform:
Statistics programming language
Data visualization tool
Open source

Ecosystem:
10,000+ packages in CRAN
Community: Scalable to big data
2.5+M users Rich application and platform integration
Taught in most universities
Popular with new and recent grads
Thriving user groups worldwide
Typical Predictive Modelling Steps
Challenges of using R
Data Movement Deployment Scale and Performance

• Moving data from the DB to R • How do I call the R script from my • Most R functions are single threaded
production application? and only accommodate datasets that
• Runtime becomes painful as data fit into available memory
volumes grow
• Movement carries security risks
ML Operationalization pre SQL 2016
Separate Service or Embedded Logic

Analytic Server
Before SQL 2016: Messy!
.. and Bandwidth issue …
In-database advanced analytics
Pushing intelligence to where data lives

Application Application

Intelligence

Intelligence
Database Database

Before Intelligence built in to the DB


SQL Server R Services solves problems
Reduce or eliminate Deploy R scripts Achieve enterprise scale
data movement with and models and performance
in-database analytics

• Leverage built-in extensibility • Use familiar T-SQL stored procedures • Use parallelism query capabilities of
mechanisms to allow secure execution to invoke R scripts from your app in-memory and ColumnStore indexes
of R scripts
• Embed the returned predictions • Leverage RevoScaleR support for
and plots large datasets and parallel algorithms
Deploy predictive analytics
Make your apps
intelligent by consuming
Use PREDICT
Deploy with to natively predictions
sp_execute_external_
predict with the model
Train a model with script and R code to
sp_execute_external_ predict with the model
Develop, explore and script and save in DB
experiment in your
favorite R IDE

Develop Train Deploy Consume


SQL Server 2016 = clean
ML Operationalization with SQL 2016

*Note:
SQL 2016: use sp_execute_external_script
SQL 2017: use PREDICT
sp_execute_external_script Patterns
Demo: ML in SQL Server
Toppings on your Frozen Yoghurt?
Like Toppings on your Frozen Yoghurt?
For demo ☺
[email protected]

#SQLFroyo

Attach Picture
Architecture

Microsoft
Flow

PowerApps
3. Tips & Tricks
SQL Server ML Features
Custom Reports for SSMS
Use SSMS custom reports from SQL Server Samples
github
SQL Server Extensibility Framework
Host external runtimes securely on SQL Server machine
Resource governance on external processes
EXTERNAL RESOURCE POOL to control CPU, Memory, CPU Affinity
Integrate with SQL query execution
New external script operator to exchange data / parameters
Parallel query pushing data to multiple external processes / threads
Streaming mode execution
Batch mode execution (in SQL Server 2017)
Implied Authentication
Impersonation for loopback connections from external scripts
Just use trusted connection in connection string
Native Scoring – New in SQL Server 2017

Native Predict function in-DB to achieve


better performance. No need to have ML Syntax sample:
Services installed! SELECT t.*, p.Label_prob as Score
FROM PREDICT(MODEL = @model,
DATA = t) WITH(Label_prob float) as p
WHERE p.Label_prob > .8
2017
*NEW*
Native Scoring for Microsoft R Server Models
T-SQL PREDICT function
In-database Package Management
CREATE EXTERNAL LIBRARY DDL
R Services in Azure SQLDB
Native Scoring using PREDICT function
Requirements:
rx* models only
Serialized model from rxSerializeModel (R)
Serialized model from rx_serialize_model (Python)

Key Benefits:
Runs natively in SQL Server (No R / Python dependency)
Low latency for execution
Ideal for highly concurrent scoring of few rows
Can be used in INSERT/UPDATE/MERGE statement directly
Don’t Do
Run R / Python script as-is Develop/Test from RTVS, PTVS,
Embed secrets in scripts RStudio or other IDE
Do data transformations that can SQL Compute Context from
be achieved in SQL client
Access network resources Data processing &
Process/transform files as part of transformations in SQL Server
the stored procedure call Data integration using SQL
Embed the R/Python code Server features
directly in applications Model management in database
4. Resources
Call to action
Resources
Getting started SQL + ML tutorials: http://aka.ms/sqldev
SSMS Reports for R Services: http://bit.ly/2r525gu
Build 2017: Serving AI with Data (SQL Server 2017):
http://bit.ly/2rbEDlZ
SQL Server Samples on GitHub:
R Services http://bit.ly/2s7tBNV
ML Services http://bit.ly/2rbLNGz
Using RTVS: http://bit.ly/2rbrXLU
Build 2017 session recordings on Channel 9: http://bit.ly/2puyVun
Microsoft Virtual Academy online: http://aka.ms/mva
Questions?
[email protected]
@MsSQLGirl
Spokane
SQL User Group
http://spokanewa.pass.org/

You might also like