46
Copyright © SAS Institute Inc. All rights reserved. Artificial Intelligence Steve Holder

Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

  • Upload
    others

  • View
    3

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Artificial IntelligenceSteve Holder

Page 2: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Artificial Intelligence

AlgorithmsMachine Learning

Deep Learning

Automation

Natural Language Processing

Computer Vision

Cognitive Computing

Page 3: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Artificial Intelligenceis the science of training systems to emulate human tasks

through learning and automation.

Page 4: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

AI and Machine LearningThere is a little Hype

Machine Learning is at the “Peak of Inflated

Expectations”

Right next to:

• Blockchain

• Cognitive Experts

• Software Defined Security

• Autonomous vehicles

Page 5: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

1950s 1980s 2010’s Present

Artificial Intelligence

Neural NetworkStatistics

Machine Learning

Machine Learning

Text, Image, Speech

Cognitive Computing

Deep Networks

Deep Learning

MACHINE LEARNING AND AI

Natural Language Processing

Page 6: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Databases

Statistics

KDD

AI

Computational Neuroscience

Data Mining

Data Science

MachineLearning

PatternRecognition

Related Disciplines

Page 7: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

How Machines Learn

▪ Machines use algorithms to:

- Study data to detect patterns.

- Apply defined rules to achieve a specified outcome.

▪ As new data is provided, the machine’s performance improves.

▪ Leading to increased or improved “intelligence.”

Page 8: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Machine Learning and AI

▪ Categorize or catalog like people or things.

▪ Predict likely outcomes or actions based on identified patterns.

▪ Identify hitherto unknown patterns and relationships.

▪ Detect anomalous or unexpected behaviors.

Page 9: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Why Now?

Data Computing Power Algorithms

We have data to find new signals

• Lower cost data. • Organizations can store

it all.• Analyst can leverage

image and text data.

We can actually process the math now

• Advanced computing technology is commoditized.

• The algorithms love it.

Refined, powerful algorithms

• Machine Learning algorithms are understood.

• Better able to automate.

Page 10: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

So what’s the difference?It’s about the process

Traditional AnalyticsHumans know best

• Assumption based• Easy to interpret• Outcome is known• Fewer variables, less data• Humans do the work:

• Select the inputs• Craft the outputs

Machine Learning Machines are smarter

• Fewer assumptions• Harder to interpret• Outcome can be unknown• Larger problems, variables and data• Machines do the work:

• Create the data• Determine the algorithms

• T-tests, ANOVA, GLMs

• MIXED models • No Variable

Selection

• Stepwise methods

• Data partitioning • Tune models on

validation data• LARS, LASSO

• Decision Trees• Bagging &

Boosting• Ensemble

Models

• Random Forests• Gradient Boosting • Neural Networks• Support Vector

Machines

• Deep Learning Neural Networks

• Hyper-Parameter Auto-Tuning

Page 11: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Types of Machine Learning Methods

SUPERVISED LEARNING

• Labeled data • Classification, Prediction• Algorithms: Logistic

Regression, Gradient Boosting etc.

UNSUPERVISED LEARNING

• Unlabeled data• Clustering, Feature

Extraction• Algorithms: K-means

clustering, PCA, etc.

SEMI-SUPERVISED LEARNING

• Labeled and unlabeled data

• Classification, Prediction• Algorithms:

Autoencoders, TSVM etc.

REINFORCEMENT LEARNING

• Agent, environment and actions

• Robotics, Gaming and Navigation

• Monte-Carlo methods etc.

Page 12: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

SAS PlatformMark Miskiman

Page 13: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Page 14: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

The Analytic Process

Page 15: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

The Analytic Process

Foundation for Analytics• Access - All sources• Build – Merge and transform• Profile – Explore your data• Fix Quality - Standardize• Governance and Lineage

Creation of Analytic Assets• Centrally Manage • Deploy in database• Create Efficiencies• Control & Adjust

Broad Range of Capabilities• Report – BI• Discover – Data Visualization• Forecast – Time Series• Optimize – Operations Research • Explain – Statistics • Predict – Data Mining, Text

Analytics, Machine Learning & AI

ModelAccess Explore AnalyzeCleanse MonitorIntegrate Govern Embed

DISCOVERYDATA DEPLOYMENT

AUTOMATE & ORCHESTRATE

Page 16: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

The Analytic Process

Broad Range of Capabilities• Report – BI• Discover – Data Visualization• Forecast – Time Series• Optimize – Operations Research • Explain – Statistics • Predict – Data Mining, Text

Analytics, Machine Learning & AI

Foundation for Analytics• Access - All sources• Build – Merge and transform• Profile – Explore your data• Fix Quality - Standardize• Governance and Lineage

Creation of Analytic Assets• Centrally Manage • Deploy in database• Create Efficiencies• Control & Adjust

Access Cleanse MonitorIntegrate Govern Embed

DISCOVERYDATA DEPLOYMENT

Analyze

Optimize

Forecast

Explore

Discover

Report

Model

Predict

Explain

Build Views

Connect

Transform

Join

Standardize

ProfileCentrally Manage

Create Efficiencies

Control & Adjust

AUTOMATE & ORCHESTRATE

Page 17: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

SAS for DataProducing data that cultivates analytics assets

17

ACCESSAccess data & assess the data structure and content.

• Browse and search available

data & metadata.

• Filter, copy & profile data.

• Lift data into memory.

• Access new data sources (files,

social media, data connections).

INTEGRATESelect data of interest, manipulate & structure it for analysis.

GOVERNAutomate data preparation tasks, monitor jobs & share plans across users.

CLEANSEPut data into a consistent, trusted format.

• Select columns & apply filters.

• Calculate Columns.

• Convert data types.

• Append, join & transpose data.

• Insert code (DATASTEP, CASL).

• Create jobs.

• Parse

• Field Extraction

• Standardize

• Match

• Identification Analysis

• Change Case

• Gender Analysis

• Manager user permissions.

• Automate jobs, view code and

logs.

• Share tables and data preparation

plans across users.

• Open APIs.

Page 18: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

SAS for DiscoveryBuilding analytics that answer any question

18

DATA DISCOVERYExplore, report, collaborate and share.

FORECASTINGLeverage time series data to forecast future results

DATA MININGExplore, find and predict patterns in data.

STATISTICSUnderstand and explain patterns in data.

• Time series forecasting.

• Simultaneous scenario

comparison.

• AI enabled forecasting

tournaments.

• Hierarchical forecasting with

reconciliation.

• Human guided algorithm assisted

discovery & modeling

• Algorithm-driven variable selection

• Deeper statistical methods for

bigger, more complex data

• Model assessments, comparisons

& tournaments on hold-out data

• Interactive Ad hoc exploration

of data.

• Discover relationships, trends

and outliers.

• Create reports and

dashboards.

• Distribute and share content

with others.

• Linear & logistic regression

• Survival analysis

• Generalized linear model, Mixed

Models

• Classification trees

• Clustering

• Fit Distributions

Page 19: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

• Unsupervised- and supervised-

learning algorithms

• Build machine learning

pipelines

• Automated model tuning

• Model assessment

comparisons & tournaments,

and scoring

MACHINE LEARNINGAutomate discovery and modeling, prediction & deployment.

SAS for DiscoveryBuilding analytics that answer any question

19

• Optimize outcome based on

constraints.

• Use cases include:

• Optimize scheduling & routing

• Site selection

• Revenue management

• Product stocking, bundling

TEXT MININGExplain & Predict patterns in unstructured text data.

• Text Analytics pipeline for

modeling

• Content Parsing, Categorization

and Analysis.

• Information retrieval, word

searches

• Word associations and linkages

• Stylometry & Interpret natural

language

OPTIMIZATIONFind optimal solutions given constraints.

Page 20: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

SAS for DeploymentRealize the value from analytics assets in action

20

GOVERNProvide oversight and governed analytics.

EMBEDEmbed and execute models, leverage compute platforms.

AUTOMATEMove from insight to action.

MONITORGain visibility into analytic outcomes and approach

• Execute models in database &

Hadoop.

• No Language Conversion.

• Generate score code for

complex models.

• Automate Model execution.

• Automate Model retraining.

• Create AI enabled model

tournaments.

• Central model management.

• Repository for SAS Models and

Open Source.

• Model and data History,

version control.

• Model and data lineage with

governance.

• Model performance reporting.

• Compare multiple models.

• Assess model accuracy (Lift, ROC,

K-S)

• Champion/Challenger modeling

• Model retraining including open

source

Page 21: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Pragmatic AI Steve Holder

Page 22: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

What people expect from AI?

Better Results Auto Magic Lower Cost Human Interfaces

• Less assumptions• Better math• More data

• Throw data at it• Remove Bias• Automate the

process

• Machines do the work

• Model tournament

• Better UI• Better interaction• Add speech, text

and images

Page 23: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

We provide elements and capabilities

for people that aspire to build an AI system.

Page 24: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

What falls under AI?

Chatbots

Smart Q&A and Search System

Deep Learning AlgorithmsRoboadvisor

Natural Language Processing

Self Learning Models

Reinforcement Algorithms

Machine LearningFacial recognition

Conversational system

CAPABILITIESAPPLICATIONS

Page 25: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

MACHINE LEARNING AND AI

25

ALGORITHMS

• Deliver robust analytic algorithms.

• Classic Statistics, Machine Learning and Deep Learning.

• More accurate analytics.

AUTOMATION

• Embed AI in SAS

• Automate the analytics lifecycle.

• Generate model tournaments for best results.

• Retrain models when they decay.

TEXT AND IMAGE PROCESSING

• Image processing, classification and recognition.

• Text processing, Analytics, & Sentiment Analysis.

• Speech to Text for SAS

• Natural language generation.

AI APPLICATIONS

• Create AI applications

• Natural language interfaces to SAS.

Page 26: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Data Manipulation

Data Manipulation

• In-Memory Data Manipulation• Frequency / Crosstab• Data Transpose• Variable Binning • Variable Cardinality Analysis• Variable Summary• Sampling and Partitioning• Missing Value Imputation• Variable Selection • Model Assessment• DS2• FedSQL• Image Processing

Page 27: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Analytics

Statistics

• Cox Proportional Hazards• Decision Trees• Design Matrix• General Additive Models• Generalized Linear Models • K-means and K-modes Clustering• Linear Regression• Logistic Regression• Nonlinear Regression • Ordinary Least Squares Regression • Partial Least Squares Regression• Pearson Correlation• Quantile Regression• Shewhart Control Chart Analysis

Page 28: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Machine Learning

Machine Learning

• Bayesian Networks• Boolean Rules• Factorization Machines• Frequent Item Set Mining• Gradient Boosting• K Nearest Neighbor• Market Basket Analysis• Moving Windows PCA• Network Analytics/Community

Detection• Random Forest• Principal Component Analysis • Support Vector Data Description• Support Vector Machines• Text Mining• Variable Clustering

Page 29: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Deep Learning

Machine Learning

Deep Forward Neural Networks (DNNs)Convolutional Neural Networks (CNNs)Support VGG-like modelsSupport ResNet modelsRecurrent Neural Networks(RNNs)Support LSTM modelSupport GRU modelAutoencoders for neural networks

Image processing extensionsAugment image actionConvert image table action Match image action2D/3D medical image visualization

Page 30: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

ModelAccess Explore AnalyzeCleanse MonitorIntegrate Govern Embed

DISCOVERYDATA DEPLOYMENT

AUTOMATE & ORCHESTRATE

Automate the Analytic Process

Data AnalysisFeature

EngineeringModel Training Deploy

More Interations

Better Analytics

IncreasedProductivity

Better Insights

Page 31: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

AI Applications

Personalized Recommendation

IoT Analytics

Person/Object Identification Three-Dimensional Scans

Personalized Offers

1 2

3 4

Page 32: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Company Conf ident ia l – For Internal Use OnlyCopyr ight © SAS Inst i tute Inc . A l l r ights reserved.

Page 33: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Company Conf ident ia l – For Internal Use OnlyCopyr ight © SAS Inst i tute Inc . A l l r ights reserved.

Over 3000 IMAGES analyzed in < 10 minutesModel Accuracy 78%

Page 34: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

ScaleHyper-parameter Auto-Tuning

SAS Auto Tuning provides:

• Automated hyper-parameters search and selection

• Guided parameter selection and defaults.

• Avoids over-fitting

• More accurate models faster vs. hand-tuning

• Patented optimization algorithm to select right combination of hyperparameters

What are hyper parameter?• Machine learning models have parameters that tune the

model for results and performance.

• Data scientist try combinations of parameters to maximize results.

• This is a manual and time consuming process.

Page 35: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

x1

x2

y = f(x1) + g(x2)

Standard Grid Search

ScalePatented Optimization

x1

x2

Random Searchx1

x2

Latin Hypercube

= individual model train and assessment

• Better hyper-parameters coverage• Optimal computation time

Page 36: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

TrustAnalytic Accuracy

• Provide new machine learning algorithms in increase accuracy

- Gradient Boosting, Factorization Machines, Support Vector Machines, Principal

Component Analysis, Deep Neural Networks

• Automated Hyperparameter auto tuning

Model accuracy using

Cut off at 80% probability Cut off at 60% probability

Number of false positives prevented

False positivesNumber of false

positives preventedFalse

positives

SAS VDMML 96 out of 98 (98%) 2 114 out of 121 (94%) 7

Other tools at Media Company

168 out of 192 (88%)

24 184 out of 247 (74%) 63

Page 37: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Diversity

Visual Interfaces

Programming Interfaces

API Interfaces

Support all users

Page 38: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

DiversitySupport for all languages

Page 39: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Workers

Controller

proc print data = hmeq (obs = 10);

run;

df = s.CASTable(‘hmeq’)

df.head(10)

df <- defCasTable(s, ‘hmeq’)

head(df, 10)

[table.fetch]

table.name = “hmeq”

from = 1 to = 10

Translated Command

APIs

TrustConsistent Execution

Page 40: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Mastering CAS ActionsHow you will become great at SAS Viya Programming

Action Sets

Actions

Parameters

Options

PROC

Page 41: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

SAS for DeploymentRealize the value from analytics assets in action

41

GOVERNProvide oversight and governed analytics.

EMBEDEmbed and execute models, leverage compute platforms.

AUTOMATEMove from insight to action.

MONITORGain visibility into analytic outcomes and approach

• Execute models in database &

Hadoop.

• No Language Conversion.

• Generate score code for

complex models.

• Automate Model execution.

• Automate Model retraining.

• Create AI enabled model

tournaments.

• Central model management.

• Repository for SAS Models and

Open Source.

• Model and data History,

version control.

• Model and data lineage with

governance.

• Model performance reporting.

• Compare multiple models.

• Assess model accuracy (Lift, ROC,

K-S)

• Champion/Challenger modeling

• Model retraining including open

source

Page 42: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

ScaleLess Technical Integration

Decentralized Centralized

Page 43: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

Data Ingestion

Data Storage Data Preparation Analytics & Modelling

Inventory & Document

Deployment & Publish

Monitoring & Retraining

Open Source

Skills Required

• Source Data structures

• Sqoop

• Hadoop & HDFS• Query in SQL, HQL, Pig• Writing customized UDF in Java• Data wrangling in R or Python• Knowledge about the models being built

• Python, PySpark, Scala• Machine Learning

Algorithms• Modelling• Spark - Parallelized

Computing

• Git, • Best Practices for

Version Control

• Unix - Bash Scripting

• Docker Containers

• Dev Ops skills

How SAS can help

• Enable the business users with access to Hadoop

• SAS can execute queries in Hadoop environment and minimize data duplication

• Provide a variety of user experiences

• Leverage open programming skills

• Consistent execution• Embedded In memory• Ability to deploy

models

• Central Model Management Repository

• Inventory all models• Model history and

lineage• Version control

• Execution in Hadoop

• No recording required

• Accessible via Restful API

• Monitor model performance

• Automatic model retraining

• Out of the box Modelperformance statistics

Open Source Analytic Lifecycle

Page 44: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

ScaleDeploy models with an Analytics Store

Why do we need an analytics store?• Machine Learning models generate huge amounts of code (up to >50M)

• Gradient Boosting, Random Forests, factorization machines etc.

• Generation and execution of this code in production can be difficult to manage.

What does the SAS Astore provider?

• A SAS Astore is a binary file that captures the state of a predictive model

• It is transportable

• Can be moved between databases and Hadoop

• No SAS import/export needed

• Allows for model deployment and execution

Page 45: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Company Conf ident ia l – For Internal Use OnlyCopyr ight © SAS Inst i tute Inc . A l l r ights reserved.

Delivers capabilities to build applications

Not limited to the cloud or just one cloud

Governed processes

SAS AI Approach

Simplified end-to-end integration

Integrated analytical streaming engine

Supported globally (tech, language...)

SAS analytic standard in the enterprise

Page 46: Artificial Intelligence - Weeblyalbertaai.weebly.com/uploads/1/1/1/4/111436951/alberta_ai_event_1.… · SAS for Discovery Building analytics that answer any question 18 DATA DISCOVERY

Copyr ight © SA S Inst i tute Inc . A l l r ights reserved.

• Leader in Advanced Analytics, AI, and Data Management

• Over 35% Market share for Advanced analytics according to IDC

• More than 83,000 sites, across 149 countries

About SAS

Founded in 1976

Giving you The Power to Know®

14,000+G L O B A L

E M P L O Y E E S

325+C A N A D I A N

E M P L O Y E E S