31
#analyticsx Copyright © 2016, SAS Institute Inc. All rights reserved. Big Data Meets Advanced Analytics Concepts and Practical Examples Dr. Gerhard Svolba – SAS DACH Competence Center Analytics

Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Big Data Meets Advanced Analytics Concepts and Practical ExamplesDr Gerhard Svolba ndash SAS DACHCompetence Center Analytics

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Use Advanced Machine Learning methods to describe the relationships in your data

bull Understand specifics of complex systems by using Monte Carlo simulations

bull Run SAS High Performance Analytics procedures in distributed mode

Concepts when Handling Big Data

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Two Practical Big Data ExamplesDeep Learning with Stacked Denoising Autoencoders

Recognize Handwritten Digits

Compress information into a few variables

Monte Carlo Simulation of the Monopolyreg Board Game

Distribution of the visit frequency on the fields

Studying the profitability of different investment decisions

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull ML wird haumlufig eingesetzt wenn Die Vorhersageguumlte eines Modells wichtiger ist als die

Erklaumlrbarkeit

Traditionelle Ansaumltze ungeeignet erscheinenraquo Mehr Variablen als Beobachtungen

raquo Viele hochkorrelierte Variablen

raquo Unstrukturierte Daten

raquo Fundamental nicht-lineare Zusammenhaumlnge

bull Anwendung Mustererkennung

Anomalie-Erkennung

Machine learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Data

Min

ing

Machine Learning

TRANSDUCTION

REINFORCEMENT LEARNING

DEVELOPMENTAL LEARNING

In semi-supervised learning supervised prediction and classification algorithms are often combined with

clustering

SEMI-SUPERVISED LEARNING

Prediction and classificationClusteringEM TSVMManifoldregularization Autoencoders

Multilayer perceptronRestricted Boltzmannmachines

SUPERVISED LEARNING

RegressionLASSO regressionLogistic regressionRidge regression

Decision treeGradient boostingRandom forests

Neural networks SVMNaiumlve BayesNeighborsGaussianprocesses

UNSUPERVISEDLEARNING

A priori rulesClustering

k-means clusteringMean shift clustering Spectral clustering

Kernel densityestimationNonnegative matrixfactorizationPCA

Kernel PCASparse PCA

Singular valuedecompositionSOM

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Classic MNIST training data

bull 784 features form a 28x28 digital grid

bull Greyscale features range from 0 to 255

bull 60000 labeled training images

(785 variables including 1 nominal target)

bull 10000 unlabeled test images(784 input variables)

Handwritten Digits as Training Data

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extract a few representative features to discriminate the digits 0-9

bull Compress information of 784 variables into 2 features

bull Use a convolutional neural network (deep learning)

Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Deep-Learning using aStacked De-noising Autoencoder

h1

h2

h3

h4

h5

Partially Corrupted Input Features

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Uncorrupted Output Features Target Layer

Input Layer

Extractable Features

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Using SAS Code to Solve the ProblemNoised INPUT

(i) INPUT(t)H1 H2 H3 H4 H5

proc neural

data= autoencoderTraining dmdbcat= workautoencoderTrainingCat

performance compile details cpucount= 12 threads= yes

archi MLP hidden= 5

hidden 300 id= h1

hidden 100 id= h2

hidden 2 id= h3 act= linear

hidden 100 id= h4

hidden 300 id= h5

input corruptedPixel1-corruptedPixel400 id= i level= int std= std

target pixel1-pixel400 act= identity id= t level= int std= std

initial random= 123 prelim 10 preiter= 10

freeze h1-gth2 freeze h2-gth3 freeze h3-gth4 freeze h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

freeze i-gth1 thaw h1-gth2

train technique= congra maxtime= 129600 maxiter= 1000

freeze h1-gth2 thaw h2-gth3

train technique= congra maxtime= 129600 maxiter= 1000

freeze h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

freeze h3-gth4 thaw h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

thaw i-gth1 thaw h1-gth2 thaw h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

code file= CPathtocodesas run

i=gtH1 H1=gtH2 H2=gtH3 H3=gtH4 H4=gtH5

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Studying a certain section in detail

h52

Target 3 Target 4Target 2Target 1

h51

W51 W52 W53 W54

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Edge Weights of the 5th layer are ldquoloadedrdquo with discriminative information

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Visualization of the separation of the two middlehidden layers

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 2: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Use Advanced Machine Learning methods to describe the relationships in your data

bull Understand specifics of complex systems by using Monte Carlo simulations

bull Run SAS High Performance Analytics procedures in distributed mode

Concepts when Handling Big Data

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Two Practical Big Data ExamplesDeep Learning with Stacked Denoising Autoencoders

Recognize Handwritten Digits

Compress information into a few variables

Monte Carlo Simulation of the Monopolyreg Board Game

Distribution of the visit frequency on the fields

Studying the profitability of different investment decisions

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull ML wird haumlufig eingesetzt wenn Die Vorhersageguumlte eines Modells wichtiger ist als die

Erklaumlrbarkeit

Traditionelle Ansaumltze ungeeignet erscheinenraquo Mehr Variablen als Beobachtungen

raquo Viele hochkorrelierte Variablen

raquo Unstrukturierte Daten

raquo Fundamental nicht-lineare Zusammenhaumlnge

bull Anwendung Mustererkennung

Anomalie-Erkennung

Machine learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Data

Min

ing

Machine Learning

TRANSDUCTION

REINFORCEMENT LEARNING

DEVELOPMENTAL LEARNING

In semi-supervised learning supervised prediction and classification algorithms are often combined with

clustering

SEMI-SUPERVISED LEARNING

Prediction and classificationClusteringEM TSVMManifoldregularization Autoencoders

Multilayer perceptronRestricted Boltzmannmachines

SUPERVISED LEARNING

RegressionLASSO regressionLogistic regressionRidge regression

Decision treeGradient boostingRandom forests

Neural networks SVMNaiumlve BayesNeighborsGaussianprocesses

UNSUPERVISEDLEARNING

A priori rulesClustering

k-means clusteringMean shift clustering Spectral clustering

Kernel densityestimationNonnegative matrixfactorizationPCA

Kernel PCASparse PCA

Singular valuedecompositionSOM

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Classic MNIST training data

bull 784 features form a 28x28 digital grid

bull Greyscale features range from 0 to 255

bull 60000 labeled training images

(785 variables including 1 nominal target)

bull 10000 unlabeled test images(784 input variables)

Handwritten Digits as Training Data

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extract a few representative features to discriminate the digits 0-9

bull Compress information of 784 variables into 2 features

bull Use a convolutional neural network (deep learning)

Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Deep-Learning using aStacked De-noising Autoencoder

h1

h2

h3

h4

h5

Partially Corrupted Input Features

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Uncorrupted Output Features Target Layer

Input Layer

Extractable Features

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Using SAS Code to Solve the ProblemNoised INPUT

(i) INPUT(t)H1 H2 H3 H4 H5

proc neural

data= autoencoderTraining dmdbcat= workautoencoderTrainingCat

performance compile details cpucount= 12 threads= yes

archi MLP hidden= 5

hidden 300 id= h1

hidden 100 id= h2

hidden 2 id= h3 act= linear

hidden 100 id= h4

hidden 300 id= h5

input corruptedPixel1-corruptedPixel400 id= i level= int std= std

target pixel1-pixel400 act= identity id= t level= int std= std

initial random= 123 prelim 10 preiter= 10

freeze h1-gth2 freeze h2-gth3 freeze h3-gth4 freeze h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

freeze i-gth1 thaw h1-gth2

train technique= congra maxtime= 129600 maxiter= 1000

freeze h1-gth2 thaw h2-gth3

train technique= congra maxtime= 129600 maxiter= 1000

freeze h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

freeze h3-gth4 thaw h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

thaw i-gth1 thaw h1-gth2 thaw h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

code file= CPathtocodesas run

i=gtH1 H1=gtH2 H2=gtH3 H3=gtH4 H4=gtH5

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Studying a certain section in detail

h52

Target 3 Target 4Target 2Target 1

h51

W51 W52 W53 W54

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Edge Weights of the 5th layer are ldquoloadedrdquo with discriminative information

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Visualization of the separation of the two middlehidden layers

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 3: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Two Practical Big Data ExamplesDeep Learning with Stacked Denoising Autoencoders

Recognize Handwritten Digits

Compress information into a few variables

Monte Carlo Simulation of the Monopolyreg Board Game

Distribution of the visit frequency on the fields

Studying the profitability of different investment decisions

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull ML wird haumlufig eingesetzt wenn Die Vorhersageguumlte eines Modells wichtiger ist als die

Erklaumlrbarkeit

Traditionelle Ansaumltze ungeeignet erscheinenraquo Mehr Variablen als Beobachtungen

raquo Viele hochkorrelierte Variablen

raquo Unstrukturierte Daten

raquo Fundamental nicht-lineare Zusammenhaumlnge

bull Anwendung Mustererkennung

Anomalie-Erkennung

Machine learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Data

Min

ing

Machine Learning

TRANSDUCTION

REINFORCEMENT LEARNING

DEVELOPMENTAL LEARNING

In semi-supervised learning supervised prediction and classification algorithms are often combined with

clustering

SEMI-SUPERVISED LEARNING

Prediction and classificationClusteringEM TSVMManifoldregularization Autoencoders

Multilayer perceptronRestricted Boltzmannmachines

SUPERVISED LEARNING

RegressionLASSO regressionLogistic regressionRidge regression

Decision treeGradient boostingRandom forests

Neural networks SVMNaiumlve BayesNeighborsGaussianprocesses

UNSUPERVISEDLEARNING

A priori rulesClustering

k-means clusteringMean shift clustering Spectral clustering

Kernel densityestimationNonnegative matrixfactorizationPCA

Kernel PCASparse PCA

Singular valuedecompositionSOM

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Classic MNIST training data

bull 784 features form a 28x28 digital grid

bull Greyscale features range from 0 to 255

bull 60000 labeled training images

(785 variables including 1 nominal target)

bull 10000 unlabeled test images(784 input variables)

Handwritten Digits as Training Data

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extract a few representative features to discriminate the digits 0-9

bull Compress information of 784 variables into 2 features

bull Use a convolutional neural network (deep learning)

Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Deep-Learning using aStacked De-noising Autoencoder

h1

h2

h3

h4

h5

Partially Corrupted Input Features

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Uncorrupted Output Features Target Layer

Input Layer

Extractable Features

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Using SAS Code to Solve the ProblemNoised INPUT

(i) INPUT(t)H1 H2 H3 H4 H5

proc neural

data= autoencoderTraining dmdbcat= workautoencoderTrainingCat

performance compile details cpucount= 12 threads= yes

archi MLP hidden= 5

hidden 300 id= h1

hidden 100 id= h2

hidden 2 id= h3 act= linear

hidden 100 id= h4

hidden 300 id= h5

input corruptedPixel1-corruptedPixel400 id= i level= int std= std

target pixel1-pixel400 act= identity id= t level= int std= std

initial random= 123 prelim 10 preiter= 10

freeze h1-gth2 freeze h2-gth3 freeze h3-gth4 freeze h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

freeze i-gth1 thaw h1-gth2

train technique= congra maxtime= 129600 maxiter= 1000

freeze h1-gth2 thaw h2-gth3

train technique= congra maxtime= 129600 maxiter= 1000

freeze h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

freeze h3-gth4 thaw h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

thaw i-gth1 thaw h1-gth2 thaw h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

code file= CPathtocodesas run

i=gtH1 H1=gtH2 H2=gtH3 H3=gtH4 H4=gtH5

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Studying a certain section in detail

h52

Target 3 Target 4Target 2Target 1

h51

W51 W52 W53 W54

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Edge Weights of the 5th layer are ldquoloadedrdquo with discriminative information

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Visualization of the separation of the two middlehidden layers

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 4: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull ML wird haumlufig eingesetzt wenn Die Vorhersageguumlte eines Modells wichtiger ist als die

Erklaumlrbarkeit

Traditionelle Ansaumltze ungeeignet erscheinenraquo Mehr Variablen als Beobachtungen

raquo Viele hochkorrelierte Variablen

raquo Unstrukturierte Daten

raquo Fundamental nicht-lineare Zusammenhaumlnge

bull Anwendung Mustererkennung

Anomalie-Erkennung

Machine learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Data

Min

ing

Machine Learning

TRANSDUCTION

REINFORCEMENT LEARNING

DEVELOPMENTAL LEARNING

In semi-supervised learning supervised prediction and classification algorithms are often combined with

clustering

SEMI-SUPERVISED LEARNING

Prediction and classificationClusteringEM TSVMManifoldregularization Autoencoders

Multilayer perceptronRestricted Boltzmannmachines

SUPERVISED LEARNING

RegressionLASSO regressionLogistic regressionRidge regression

Decision treeGradient boostingRandom forests

Neural networks SVMNaiumlve BayesNeighborsGaussianprocesses

UNSUPERVISEDLEARNING

A priori rulesClustering

k-means clusteringMean shift clustering Spectral clustering

Kernel densityestimationNonnegative matrixfactorizationPCA

Kernel PCASparse PCA

Singular valuedecompositionSOM

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Classic MNIST training data

bull 784 features form a 28x28 digital grid

bull Greyscale features range from 0 to 255

bull 60000 labeled training images

(785 variables including 1 nominal target)

bull 10000 unlabeled test images(784 input variables)

Handwritten Digits as Training Data

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extract a few representative features to discriminate the digits 0-9

bull Compress information of 784 variables into 2 features

bull Use a convolutional neural network (deep learning)

Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Deep-Learning using aStacked De-noising Autoencoder

h1

h2

h3

h4

h5

Partially Corrupted Input Features

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Uncorrupted Output Features Target Layer

Input Layer

Extractable Features

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Using SAS Code to Solve the ProblemNoised INPUT

(i) INPUT(t)H1 H2 H3 H4 H5

proc neural

data= autoencoderTraining dmdbcat= workautoencoderTrainingCat

performance compile details cpucount= 12 threads= yes

archi MLP hidden= 5

hidden 300 id= h1

hidden 100 id= h2

hidden 2 id= h3 act= linear

hidden 100 id= h4

hidden 300 id= h5

input corruptedPixel1-corruptedPixel400 id= i level= int std= std

target pixel1-pixel400 act= identity id= t level= int std= std

initial random= 123 prelim 10 preiter= 10

freeze h1-gth2 freeze h2-gth3 freeze h3-gth4 freeze h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

freeze i-gth1 thaw h1-gth2

train technique= congra maxtime= 129600 maxiter= 1000

freeze h1-gth2 thaw h2-gth3

train technique= congra maxtime= 129600 maxiter= 1000

freeze h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

freeze h3-gth4 thaw h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

thaw i-gth1 thaw h1-gth2 thaw h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

code file= CPathtocodesas run

i=gtH1 H1=gtH2 H2=gtH3 H3=gtH4 H4=gtH5

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Studying a certain section in detail

h52

Target 3 Target 4Target 2Target 1

h51

W51 W52 W53 W54

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Edge Weights of the 5th layer are ldquoloadedrdquo with discriminative information

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Visualization of the separation of the two middlehidden layers

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 5: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Data

Min

ing

Machine Learning

TRANSDUCTION

REINFORCEMENT LEARNING

DEVELOPMENTAL LEARNING

In semi-supervised learning supervised prediction and classification algorithms are often combined with

clustering

SEMI-SUPERVISED LEARNING

Prediction and classificationClusteringEM TSVMManifoldregularization Autoencoders

Multilayer perceptronRestricted Boltzmannmachines

SUPERVISED LEARNING

RegressionLASSO regressionLogistic regressionRidge regression

Decision treeGradient boostingRandom forests

Neural networks SVMNaiumlve BayesNeighborsGaussianprocesses

UNSUPERVISEDLEARNING

A priori rulesClustering

k-means clusteringMean shift clustering Spectral clustering

Kernel densityestimationNonnegative matrixfactorizationPCA

Kernel PCASparse PCA

Singular valuedecompositionSOM

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Classic MNIST training data

bull 784 features form a 28x28 digital grid

bull Greyscale features range from 0 to 255

bull 60000 labeled training images

(785 variables including 1 nominal target)

bull 10000 unlabeled test images(784 input variables)

Handwritten Digits as Training Data

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extract a few representative features to discriminate the digits 0-9

bull Compress information of 784 variables into 2 features

bull Use a convolutional neural network (deep learning)

Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Deep-Learning using aStacked De-noising Autoencoder

h1

h2

h3

h4

h5

Partially Corrupted Input Features

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Uncorrupted Output Features Target Layer

Input Layer

Extractable Features

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Using SAS Code to Solve the ProblemNoised INPUT

(i) INPUT(t)H1 H2 H3 H4 H5

proc neural

data= autoencoderTraining dmdbcat= workautoencoderTrainingCat

performance compile details cpucount= 12 threads= yes

archi MLP hidden= 5

hidden 300 id= h1

hidden 100 id= h2

hidden 2 id= h3 act= linear

hidden 100 id= h4

hidden 300 id= h5

input corruptedPixel1-corruptedPixel400 id= i level= int std= std

target pixel1-pixel400 act= identity id= t level= int std= std

initial random= 123 prelim 10 preiter= 10

freeze h1-gth2 freeze h2-gth3 freeze h3-gth4 freeze h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

freeze i-gth1 thaw h1-gth2

train technique= congra maxtime= 129600 maxiter= 1000

freeze h1-gth2 thaw h2-gth3

train technique= congra maxtime= 129600 maxiter= 1000

freeze h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

freeze h3-gth4 thaw h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

thaw i-gth1 thaw h1-gth2 thaw h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

code file= CPathtocodesas run

i=gtH1 H1=gtH2 H2=gtH3 H3=gtH4 H4=gtH5

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Studying a certain section in detail

h52

Target 3 Target 4Target 2Target 1

h51

W51 W52 W53 W54

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Edge Weights of the 5th layer are ldquoloadedrdquo with discriminative information

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Visualization of the separation of the two middlehidden layers

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 6: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Classic MNIST training data

bull 784 features form a 28x28 digital grid

bull Greyscale features range from 0 to 255

bull 60000 labeled training images

(785 variables including 1 nominal target)

bull 10000 unlabeled test images(784 input variables)

Handwritten Digits as Training Data

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extract a few representative features to discriminate the digits 0-9

bull Compress information of 784 variables into 2 features

bull Use a convolutional neural network (deep learning)

Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Deep-Learning using aStacked De-noising Autoencoder

h1

h2

h3

h4

h5

Partially Corrupted Input Features

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Uncorrupted Output Features Target Layer

Input Layer

Extractable Features

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Using SAS Code to Solve the ProblemNoised INPUT

(i) INPUT(t)H1 H2 H3 H4 H5

proc neural

data= autoencoderTraining dmdbcat= workautoencoderTrainingCat

performance compile details cpucount= 12 threads= yes

archi MLP hidden= 5

hidden 300 id= h1

hidden 100 id= h2

hidden 2 id= h3 act= linear

hidden 100 id= h4

hidden 300 id= h5

input corruptedPixel1-corruptedPixel400 id= i level= int std= std

target pixel1-pixel400 act= identity id= t level= int std= std

initial random= 123 prelim 10 preiter= 10

freeze h1-gth2 freeze h2-gth3 freeze h3-gth4 freeze h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

freeze i-gth1 thaw h1-gth2

train technique= congra maxtime= 129600 maxiter= 1000

freeze h1-gth2 thaw h2-gth3

train technique= congra maxtime= 129600 maxiter= 1000

freeze h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

freeze h3-gth4 thaw h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

thaw i-gth1 thaw h1-gth2 thaw h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

code file= CPathtocodesas run

i=gtH1 H1=gtH2 H2=gtH3 H3=gtH4 H4=gtH5

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Studying a certain section in detail

h52

Target 3 Target 4Target 2Target 1

h51

W51 W52 W53 W54

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Edge Weights of the 5th layer are ldquoloadedrdquo with discriminative information

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Visualization of the separation of the two middlehidden layers

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 7: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extract a few representative features to discriminate the digits 0-9

bull Compress information of 784 variables into 2 features

bull Use a convolutional neural network (deep learning)

Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Deep-Learning using aStacked De-noising Autoencoder

h1

h2

h3

h4

h5

Partially Corrupted Input Features

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Uncorrupted Output Features Target Layer

Input Layer

Extractable Features

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Using SAS Code to Solve the ProblemNoised INPUT

(i) INPUT(t)H1 H2 H3 H4 H5

proc neural

data= autoencoderTraining dmdbcat= workautoencoderTrainingCat

performance compile details cpucount= 12 threads= yes

archi MLP hidden= 5

hidden 300 id= h1

hidden 100 id= h2

hidden 2 id= h3 act= linear

hidden 100 id= h4

hidden 300 id= h5

input corruptedPixel1-corruptedPixel400 id= i level= int std= std

target pixel1-pixel400 act= identity id= t level= int std= std

initial random= 123 prelim 10 preiter= 10

freeze h1-gth2 freeze h2-gth3 freeze h3-gth4 freeze h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

freeze i-gth1 thaw h1-gth2

train technique= congra maxtime= 129600 maxiter= 1000

freeze h1-gth2 thaw h2-gth3

train technique= congra maxtime= 129600 maxiter= 1000

freeze h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

freeze h3-gth4 thaw h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

thaw i-gth1 thaw h1-gth2 thaw h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

code file= CPathtocodesas run

i=gtH1 H1=gtH2 H2=gtH3 H3=gtH4 H4=gtH5

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Studying a certain section in detail

h52

Target 3 Target 4Target 2Target 1

h51

W51 W52 W53 W54

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Edge Weights of the 5th layer are ldquoloadedrdquo with discriminative information

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Visualization of the separation of the two middlehidden layers

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 8: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Deep-Learning using aStacked De-noising Autoencoder

h1

h2

h3

h4

h5

Partially Corrupted Input Features

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Hidden Neurons

Uncorrupted Output Features Target Layer

Input Layer

Extractable Features

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Using SAS Code to Solve the ProblemNoised INPUT

(i) INPUT(t)H1 H2 H3 H4 H5

proc neural

data= autoencoderTraining dmdbcat= workautoencoderTrainingCat

performance compile details cpucount= 12 threads= yes

archi MLP hidden= 5

hidden 300 id= h1

hidden 100 id= h2

hidden 2 id= h3 act= linear

hidden 100 id= h4

hidden 300 id= h5

input corruptedPixel1-corruptedPixel400 id= i level= int std= std

target pixel1-pixel400 act= identity id= t level= int std= std

initial random= 123 prelim 10 preiter= 10

freeze h1-gth2 freeze h2-gth3 freeze h3-gth4 freeze h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

freeze i-gth1 thaw h1-gth2

train technique= congra maxtime= 129600 maxiter= 1000

freeze h1-gth2 thaw h2-gth3

train technique= congra maxtime= 129600 maxiter= 1000

freeze h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

freeze h3-gth4 thaw h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

thaw i-gth1 thaw h1-gth2 thaw h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

code file= CPathtocodesas run

i=gtH1 H1=gtH2 H2=gtH3 H3=gtH4 H4=gtH5

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Studying a certain section in detail

h52

Target 3 Target 4Target 2Target 1

h51

W51 W52 W53 W54

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Edge Weights of the 5th layer are ldquoloadedrdquo with discriminative information

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Visualization of the separation of the two middlehidden layers

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 9: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Using SAS Code to Solve the ProblemNoised INPUT

(i) INPUT(t)H1 H2 H3 H4 H5

proc neural

data= autoencoderTraining dmdbcat= workautoencoderTrainingCat

performance compile details cpucount= 12 threads= yes

archi MLP hidden= 5

hidden 300 id= h1

hidden 100 id= h2

hidden 2 id= h3 act= linear

hidden 100 id= h4

hidden 300 id= h5

input corruptedPixel1-corruptedPixel400 id= i level= int std= std

target pixel1-pixel400 act= identity id= t level= int std= std

initial random= 123 prelim 10 preiter= 10

freeze h1-gth2 freeze h2-gth3 freeze h3-gth4 freeze h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

freeze i-gth1 thaw h1-gth2

train technique= congra maxtime= 129600 maxiter= 1000

freeze h1-gth2 thaw h2-gth3

train technique= congra maxtime= 129600 maxiter= 1000

freeze h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

freeze h3-gth4 thaw h4-gth5

train technique= congra maxtime= 129600 maxiter= 1000

thaw i-gth1 thaw h1-gth2 thaw h2-gth3 thaw h3-gth4

train technique= congra maxtime= 129600 maxiter= 1000

code file= CPathtocodesas run

i=gtH1 H1=gtH2 H2=gtH3 H3=gtH4 H4=gtH5

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Studying a certain section in detail

h52

Target 3 Target 4Target 2Target 1

h51

W51 W52 W53 W54

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Edge Weights of the 5th layer are ldquoloadedrdquo with discriminative information

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Visualization of the separation of the two middlehidden layers

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 10: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Studying a certain section in detail

h52

Target 3 Target 4Target 2Target 1

h51

W51 W52 W53 W54

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Edge Weights of the 5th layer are ldquoloadedrdquo with discriminative information

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Visualization of the separation of the two middlehidden layers

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 11: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Edge Weights of the 5th layer are ldquoloadedrdquo with discriminative information

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Visualization of the separation of the two middlehidden layers

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 12: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Visualization of the separation of the two middlehidden layers

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 13: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Our method results in much better separation that simple principal components analysis

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 14: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Extremely accurate predictions using deep neural networks

bull bdquoTarget Variableldquo Digit 0-9 has not been used in the model

bull ldquoFeature Extractionldquo as pre-step in predictive modeling

bull Requires Model-Tuning

bull The most common applications of deep learning involve pattern recognition in unstructured data such as text photos videos and sound

Summary Semi-Supervised Learning

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 15: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

The Monopolyreg board game is a complex system

Set of Complex Rules

Additional Instructions

Framework of Opportunities and Events

Random

Components

Monetary Dimension

Dynamic

Component

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 16: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull What is the distribution of visits on the fields of the board game

bull Which fields are most profitable

bull Which fields to have a high variability in profitability

bull These questions can be transferred to many other simulations studies of complex systems

Questions of Interest

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 17: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Locating the Token ndash Influential FactorsSum of

2 Dice

Accelerator

Dice

Event Fields

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 18: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Almost Even DistributionSum of

2 Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 19: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

All Field-40 visits are relocated to 14Sum of

2 Dice

Go to Jail

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 20: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Event Fields relocate to other fieldsSum of

2 Dice

Go to Jail

Event Fields

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 21: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Red Dice introduces high variabilitySum of

2 Dice

Go to Jail

Event Fields

Accelerator

Dice

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 22: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 20 rounds

bull If the 3rd dice showsthe Monopolyreg manbull Move forward to the

next free property-field

bull The the next propertyfield otherwiese

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 23: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Effect of the accelerator dice after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 24: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Example for a Relocation

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 25: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 40 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 26: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Profitability Distribution after 70 rounds

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 27: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Implementation in SAS

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 28: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Applying advanced analytical methods to big data allows you to better understand relationshiops in the underlying processes

bull You receive results that would otherweise remainundiscovered

bull SAS offers a full set of methods to handle big data in advanced analytics applications

Summary

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 29: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

bull Patric Hall ldquoOverview of Machine Learning with SAS Enterprise Minerrdquo httpsupportsascomresourcespapersproceedings14SAS313-2014pdf

bull Rick Wicklin Simulating Data with SAS httpsupportsascompublishingauthorswicklinhtml

bull Gerhard Svolba Applying Data Science Business Case Studies Using SAS (SAS Press expected 2017)

Links

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 30: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

analyticsx

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

Contact InformationGerhard SvolbaAnalytic Solution ArchitectSAS-AustriaSastoolsbygerhardgmxnethttpwwwsascommunityorgwikiGerhard_SvolbaLinkedIn ndash XING ndash PictureBlog

Data Quality for Analytics Using SASSAS Press 2012httpwwwsascommunityorgwikiData_Quality_for_Analytics

Data Preparation for Analytics Using SASSAS Press 2006

httpwwwsascommunityorgwikiData_Preparation_for_Analytics

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx

Page 31: Big Data Meets Advanced Analytics Concepts and …...•“Feature Extraction“ as pre-step in predictive modeling •Requires Model-Tuning •The most common applications of deep

Copyr i g ht copy 2016 SAS Ins t i tu t e Inc A l l r ights reser ve d

analyticsx