SAP HANA SPS10- Predictive Analysis Library and Application Function Modeler

1© 2015 SAP SE or an SAP affiliate company. All rights reserved.
SAP HANA SPS 10 – What’s New?
Predictive Analysis Library & Application Function Modeler
SAP HANA Product Management June, 2015
(Delta from SPS 09 to SPS 10)

© 2015 SAP SE or an SAP affiliate company. All rights reserved. 2Customer
Disclaimer
This presentation outlines our general product direction and should not be relied on in making
a purchase decision. This presentation is not subject to your license agreement or any other
agreement with SAP.
SAP has no obligation to pursue any course of business outlined in this presentation or to
develop or release any functionality mentioned in this presentation. This presentation and
SAP’s strategy and possible future developments are subject to change and may be changed
by SAP at any time for any reason without notice.
This document is provided without a warranty of any kind, either express or implied, including
but not limited to, the implied warranties of merchantability, fitness for a particular purpose, or
non-infringement. SAP assumes no responsibility for errors or omissions in this document,
except if such damages were caused by SAP intentionally or grossly negligent.

Predictive Analysis Library (PAL)

Agenda
Predictive Analysis Library (PAL)
Release Theme
List of Algorithms
General Changes
New Algorithms
• Confusion Matrix
• Parameter Selection and Model Evaluation
• Gaussian Mixture Model
• Latent Dirichlet Allocation (LDA)
• Test for White Noise, Trend, Seasonality
• Grubbs Outlier Test
• Seasonal ARIMA
Enhancements
Documentation

HANA Predictive Analysis Library – What’s New in SPS 10?
Release Theme
The SPS 10 version of the Predictive Analysis Library includes many new algorithms as well as
several enhancements to existing algorithms.
These new features were chosen based on the prioritization of customer and other stakeholder
requests.

SAP HANA In-Memory Predictive Analytics
Predictive Analysis Library (PAL) - Algorithms Supported
Association Analysis
 Apriori
 Apriori Lite
 FP-Growth
 KORD – Top K Rule Discovery
Classification Analysis
 CART *
 C4.5 Decision Tree Analysis
 CHAID Decision Tree Analysis
 K Nearest Neighbour
 Logistic Regression
 Back-Propagation (Neural Network)
 Naïve Bayes
 Support Vector Machine
 Confusion Matrix*
 Parameter Selection & Model Evaluation*
Regression
 Multiple Linear Regression
 Polynomial Regression
 Exponential Regression
 Bi-Variate Geometric Regression
 Bi-Variate Logarithmic Regression
Probability Distribution
 Distribution Fit
 Cumulative Distribution Function
 Quantile Function
Outlier Detection
 Inter-Quartile Range Test (Tukey’s Test)
 Variance Test
 Anomaly Detection
 Grubbs Outlier Test*
Link Prediction
 Common Neighbors
 Jaccard’s Coefficient
 Adamic/Adar
 Katzβ
Data Preparation
 Sampling
 Random Distribution Sampling
 Binning
 Scaling
 Partitioning
 Principal Component Analysis (PCA) * New in SPS 10
Statistic Functions
(Univariate)
 Mean, Median, Variance, Standard
Deviation
 Kurtosis
 Skewness
Statistic Functions
(Multivariate)
 Covariance Matrix
 Pearson Correlations Matrix
 Chi-squared Tests:
- Test of Quality of Fit
- Test of Independence
 F-test (variance equal test)
Other
 Weighted Scores Table
 Substitute Missing Values
Cluster Analysis
 ABC Classification
 DBSCAN
 K-Means
 K-Medoid Clustering
 K-Medians
 Kohonen Self Organized Maps
 Agglomerate Hierarchical
 Affinity Propagation
 Gaussian Mixture Model*
 Latent Dirichlet Allocation (LDA)*
Time Series Analysis
 Single Exponential Smoothing
 Double Exponential Smoothing
 Triple Exponential Smoothing
 Forecast Smoothing
 ARIMA / Seasonal ARIMA*
 Brown Exponential Smoothing
 Croston Method
 Forecast Accuracy Measure
 Linear Regression with Damped Trend
and Seasonal Adjust
 Test for White Noise, Trend, Seasonality*

General Changes
Enhanced PAL exception handling for HANA SQL handler
Exceptions thrown by PAL could be caught by the exception handler in a SQLScript procedure.
Error code 423 (AFL error).
PAL function integration with SAP HANA Series Data and window functions (partition, binning,
single & double exponential smoothing)
RANDOM_PARTITION
(<training_set_size>,
<validation_set_size>, <test_set_size>,
[<seed>]) OVER ( [ PARTITION BY
<expression> [ { , <expression> } ... ] ]
ORDER BY
<window_order_by_expression> )
BINNING (VALUE => <column ref>,
<binning parameter> => <expression>)
OVER ([PARTITION BY …])
<binning parameter> := BIN_COUNT |
BIN_WIDTH | TILE_COUNT |
STDDEV_COUNT
SERIES_FILTER
(VALUE, METHOD_NAME => <NAME>,
<filter parameter> => <expression>
OVER ( PARTITION BY … ORDER BY
…)
<NAME> :=
SINGLESMOOTH/DOUBLESMOOTH
<filter parameter> := ALPHA|BETA

Parameter Selection and Model Evaluation
Parameter Selection and Model Evaluation
Enables cross-validation and parameter selection
for the following PAL functions:
• Logistic Regression
• Naive Bayes
• Support Vector Machine
To avoid over fitting and optimize model parameters, it is
common to use cross validation to evaluate model
performance and perform model selection.
This function is an envelope for different classification
algorithms to provide automatic parameter selection and
model evaluation facilities during the model training
phase.
Output Tables

Grubbs Outlier Test - Anomaly detection
Grubbs Outlier Test - Anomaly detection
Grubbs’ test is used to detect outliers using hypothesis test and Grubbs’ test statistic from a given
univariate data set , and the algorithm assumes that Y comes from Gaussian distribution.

Gaussian Mixture Model (GMM)
Gaussian Mixture Model
GMM is a probabilistic model that assumes the
data points are generated by a mixture of a
number of Gaussian distributions with unknown
parameters. It can be used to cluster data points
with the probability of belonging to each cluster.
Each component in GMM has its own weight,
mean and covariance matrix. The weight means
the importance of a Gaussian distribution in the
GMM. The mean and covariance matrix are the
basic parameters of a Gaussian distribution.
As one example, GMM can be used in the field of image segmentation
and clustering. In PAL, we view the GMM as a clustering algorithm. We
use it to describe the data and finally we can get the probabilities of a
sample belonging to each of the Gaussian components in GMM.

Latent Dirichlet Allocation (LDA) – Topic Modeling
Latent Dirichlet allocation
Latent Dirichlet allocation (LDA) is a generative model that allows sets of observations to be explained
by unobserved groups that explain why some parts of the data are similar. It is often used in text
mining for topic modeling. If observations are words collected into documents, it posits that each
document is a mixture of a small number of topics and that each word's creation is attributable to one
of the document's topics. In PAL, the parameter inference is done via Gibbs sampling.
Estimation
Inference

Confusion Matrix (CM)
Confusion Matrix
Confusion matrix is a traditional method to evaluate the performance of classification algorithms,
including multi-class problems. PAL confusion matrix calculates the precision, recall and F1-score/ Fβ-
score.
Input: ID, Original Label, Predicted Label
Output: Class, Recall, Precision, F-measure, Support
Predicted Class
Actual
Class

Test for White Noise
Test for White Noise
The algorithm is used to identify if a time series has white noise or not. If white noise exists in the raw
time series, the value of 1 will be returned by the algorithm as an indication, if not, the value of 0 will be
returned.
In PAL, Ljung-Box test is used to test for autocorrelation at different lags.
White noise refers to the signal power distribution being independent over time or among frequencies.
1 = White noise exists
0 = White noise does
NOT exist

Test for Trend
Test for Trend
The algorithm is to identify if a time series has a trend, i.e., upward, downward, or no trend, and
calculate the de-trended time series. Two methods are provided for identifying the trend, which are the
difference-sign test and the rank test.

Test for Seasonality
Test for Seasonality – identify if seasonality exists for time series
The algorithm is to test if a time series has a seasonality or not. If it does, the corresponding
seasonality model (additive or multiplicative) is identified, and also the de-seasonalized series (i.e.,
both trend and seasonality are eliminated) is given.

Enhancements (1 of 4)
Random Distribution Sampling
• Triangular distribution support
Logistic Regression
• Added Elastic Net regularization (linear combination of L1 and L2 regularization) to optimize fitting
of a model
• Add cancel flag
Multiple Linear Regression
• Change default optimization method to QR decomposition
• Add cancel flag
• Added Elastic Net regularization (linear combination of L1 and L2 regularization) to optimize fitting
of a model

Decision Trees (C4.5, CART, CHAID)
• Enable column selection
• Support asymmetric class by allowing penalty on misclassification of certain classes
ARIMA
• Support S-ARIMA, leverage seasonality patterns in ARIMA, ARIMA-X
• Improved algorithm performance
Exponential Smoothing
• Added dampened trend to double and triple exponential smoothing algorithms. PAL provides two
types of double exponential smoothing: Holt Linear Exponential Smoothing and Additive Damped
Trend Holt Linear Exponential Smoothing. Holt’s linear method displays a constant trend indefinitely
into the future. Empirical evidence indicates that the Holt linear method tends to over-forecast.
Hence a parameter used to damp the trend may improve the situation. For additive triple
exponential smoothing, one additive damped method is also supported.

Substitute missing values function
• Additional substitution strategy to replace the missing values with zeroes and specified value
Apriori
• Add time-out capability
• Remove the data type restriction of rule output for APRIORI2
AprioriLite
• Introduce a time-out in the algorithm
FP-Growth
• Provide relational table output for generated rules
• Introduce a time-out in the algorithm, add cancel flag
• Add extra filter parameters like LHS, RHS, MAX_CONSEQUENT
• Add parallelism for rule generation

Neural Network BP
• Add multiple thread support for batch training
• Variable learning rate for stochastic training
• Enhanced stochastic decent optimization
• Added multiple thread support for prediction
• Added SoftMax output parameter for prediction
SVM
• Support categorical variables
• Enhance output tables with more statistics
DBSCAN
• Support categorical variable

How to find SAP HANA documentation on this topic?
SAP HANA Platform SPS
 What’s New – Release Notes
 Installation
– SAP HANA Server InstallationGuide
 Administration
– SAP HANA Administration Guide
 SAP HANA Options
– SAP HANA Predictive Analysis Library (PAL) Reference
 Development
– SAP HANA Developer Guide
• In addition to this learning material, you can find SAP HANA documentation on
SAP Help Portal knowledge center at http://help.sap.com/hana_platform.
• The knowledge center is structured according to the product lifecycle: installation, security, administration,
development. So you can find e.g. the SAP HANA Predictive Analysis Library (PAL) Reference
in the SAP HANA Optionst section and so forth …

Web-Based Application Function Modeler for SAP HANA
Graphical Re-Design and DataFlow Enhancements with HANA SPS10
New Web-based Flowgraph
Editor
• Support opening flowgraph file
• Support for AFL transform
• Support for R-Script tranform
• Support for SDI/SDQ
• Create Procedure or Task
runtime options
Graphical
dataflow
modeling
Compose Application
Function Calls (PAL,
BFL, …)
Writing custom
operators in R
+Information
Management &
DataQuality Operations
Standard Set-
Operation+
+
• Interoperability of Flowgraph editors -
Flowgraph created in web-based editor
works in Hana Studio editor and vice versa.
Note: Not all AFL functions will be supported in
first SPS10 revision. PAL is NOT supported in
first SPS10 revision.

Web-Based Application Function Modeler for SAP HANA
AFL Node enables selection of function by area
• AFL functions do not individually
appear in the palette as they do in
SAP HANA studio.
• One AFL Node can be selected
• The AFL function is then selected.
Note: Not all AFL functions are
supported in initial SPS10 revision.
PAL is NOT supported in first SPS10
revision.

Application Function Modeler Enhancements
SPS10
General
• Support opening empty flowgraph file
• Update the regular expression in existing
PAL palette template to fit PAL SPS10
• Enhanced flow validation, NCLOB support
Added new PAL functions into the palette
template (9 new base functions)
• Time Series: Trend Test, White Noise Test,
Seasonality Test
• Clustering: GMM, LDA Estimate, LDA
Inference
• Classification: PSME (Parameter selection
and model evaluation), Confusion Matrix
• Statistics: Grubbs Test
Update the function parameters in existing PAL
palette template
• Remove deprecated parameters
• Add new available parameters
• Update the default value for some existing
parameters
Added PAL Overload Functions (14 functions)
Clustering
• ANOMALYDETECTION__OVERLOAD_2_3
• KMEANS__OVERLOAD_2_4
• LDAESTIMATE__OVERLOAD_2_6
• LDAINFERENCE__OVERLOAD_5_2
• LDAINFERENCE__OVERLOAD_5_3
Time Series
• SINGLESMOOTH__OVERLOAD_2_2
• DOUBLEMOOTH__OVERLOAD_2_2
• TRIPLESMOOTH__OVERLOAD_2_2
Note: the function <Base Name>__OVERLOAD_2_3 is the overload function with 2
input tables and 3 output tables.
Classification
• PREDICTWITHBPNN__OVERLOA
D_3_2
• PSME__OVERLOAD_3_4
• PSME__OVERLOAD_3_5
• SVMPREDICT__OVERLOAD_5_1
• SVMTRAIN__OVERLOAD_2_3
Association
• FPGROWTH__OVERLOAD_2_3

© 2015 SAP SE or an SAP affiliate company. All rights reserved. 31Public
How to find SAP HANA documentation on this topic?
SAP HANA Platform documentation
 What’s New – Release Notes
 Modeling Information
– SAP HANA Modeling Guide
 Development Information
– SAP HANA Developer Guide
 References
– SAP HANA SQL Reference
• In addition to this learning material, you find SAP HANA documentation on
SAP Help Portal knowledge center at
http://help.sap.com/hana_platform.
• The knowledge center is structured according to the product lifecycle:
installation > security > administration > modeling > development.
So you can find e.g. the SAP HANA Developer Guide
in the Develepment section and so forth …

How to find SAP HANA demo examples on this topic?
• Go Online under https://www.youtube.com/user/saphanaacademy .

© 2015 SAP SE or an SAP affiliate company. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or for any purpose without the express permission of SAP SE or an SAP affiliate company.
SAP and other SAP products and services mentioned herein as well as their respective logos are trademarks or registered trademarks of SAP SE (or an SAP affiliate
company) in Germany and other countries. Please see http://global12.sap.com/corporate-en/legal/copyright/index.epx for additional trademark information and notices.
Some software products marketed by SAP SE and its distributors contain proprietary software components of other software vendors.
National product specifications may vary.
These materials are provided by SAP SE or an SAP affiliate company for informational purposes only, without representation or warranty of any kind, and SAP SE or its
affiliated companies shall not be liable for errors or omissions with respect to the materials. The only warranties for SAP SE or SAP affiliate company products and
services are those that are set forth in the express warranty statements accompanying such products and services, if any. Nothing herein should be construed as
constituting an additional warranty.
In particular, SAP SE or its affiliated companies have no obligation to pursue any course of business outlined in this document or any related presentation, or to develop
or release any functionality mentioned therein. This document, or any related presentation, and SAP SE’s or its affiliated companies’ strategy and possible future
developments, products, and/or platform directions and functionality are all subject to change and may be changed by SAP SE or its affiliated companies at any time
for any reason without notice. The information in this document is not a commitment, promise, or legal obligation to deliver any material, code, or functionality. All forward-
looking statements are subject to various risks and uncertainties that could cause actual results to differ materially from expectations. Readers are cautioned not to place
undue reliance on these forward-looking statements, which speak only as of their dates, and they should not be relied upon in making purchasing decisions.

Planned Innovations Future DirectionTODAY
SAP HANA Predictive Analysis Library Roadmap
Product road map overview – key themes and capabilities
Framework
 SAP HANA built-in C++ library for Advanced Analytics
 Accessible from Application Function Layer via
SQLScript, SAP Predictive Analysis, and Application
Function Modeler
Algorithms and Functions
 Regression
Multiple Linear Regression, Polynomial / Exponential / Bi-
Variate Geometric / Logarithmic Regression
 Classification
(Multi-class) Log. Regression, SVM, C4.5 & CHAID &
CART Dec. Trees, KNN , Naive Bayes, Neural Network…
 Clustering
K-means/median/Medoid/, DBSCAN, Hierarchical
clustering, Affinity Propagation…
 Association Rule Learning
Apriori / Apriori Lite, FP-Growth
 Time Series Analysis
Exponential/AHEAD smoothing, ARIMA-X, Croston
method
 Others
Distribution fit, Random sampling, Link prediction, ABC
analysis, Univariate / multivariate statistics…
Framework
 Enhanced PAL exception handling for HANA
SQL handler
 Disable L Interface – except for Apriori, Apriori
Lite, and Logistic Regression
 PAL function integration with SAP HANA Series
Data and window function (single & double
exponential smoothing, binning, partion)
Algorithms
 Topic model algorithm (Latent Dirichlet allocation
LDA)
 Time series enhancement (Seasonal-ARIMA,&
ARIMA-X, test of trend/seasonality/white noise,
…)
 Grubbs Outlier Test - Anomaly detection
 Gaussian Mixture Model (clustering with
probability)
 Confusion Matrix - Classification evaluation
 Enable cross-validation for some PAL functions –
Naïve Bayes, SVM and Logistic Regression
 Various enhancements to many algorithms
Framework
 Generalized storage and consumption of
predictive models
 Simplified programming approach
 Architecture/Interface improvements
 Enable PAL functions in distributed HANA
environment
 Built-in PAL functions in HANA without
explicit AFL installation
Algorithms
 Meta-heuristic optimization
 Unified linear model (Generalized linear
model, Generalized Additive Model)
 Sparse data analysis
 Tests of goodness of fit enhancements
 Random Forest
 Recommendation algorithms
 Discriminant analysis
 Bayesian optimization
 Matrix factorization
 …
This is the current state of planning and may be changed by SAP at any time.(Release SPS09) (To be released in SPS10)

SAP HANA SPS10- Predictive Analysis Library and Application Function Modeler

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (20)

En vedette

En vedette (19)

Similaire à SAP HANA SPS10- Predictive Analysis Library and Application Function Modeler

Similaire à SAP HANA SPS10- Predictive Analysis Library and Application Function Modeler (20)

Plus de SAP Technology

Plus de SAP Technology (20)

Dernier

Dernier (20)

SAP HANA SPS10- Predictive Analysis Library and Application Function Modeler