data science amsterdam - massively parallel processing with procedural languages

BUILT FOR THE SPEED OF BUSINESS

Additional Line 18 Point Verdana

@ianhuston 2 © Copyright 2014 Pivotal. All rights reserved. 2 © Copyright 2013 Pivotal. All rights reserved. 2 © Copyright 2013 Pivotal. All rights reserved. 2 © Copyright 2013 Pivotal. All rights reserved. 2 © Copyright 2014 Pivotal. All rights reserved.

Massively Parallel Processing with Procedural Languages

PL/R and PL/Python

Ian Huston, @ianhuston Data Scientist, Pivotal

@ianhuston 3 © Copyright 2014 Pivotal. All rights reserved.

Some Links for this talk

� Simple code examples: https://github.com/ihuston/plpython_examples

�  IPython notebook rendered with nbviewer: http://tinyurl.com/ih-plpython

� More info (especially for PL/R): http://gopivotal.github.io/gp-r/


The Pivotal Message

Agile: data driven apps and rapid time to value Data Lake: store everything, analyze anything Enterprise PaaS: revolutionary software development and speed; build the right thing


Open Source is Pivotal


Introducing Pivotal Data Labs Our Charter: Pivotal Data Labs is Pivotal’s differentiated and highly opinionated data-centric service delivery organization.

Our Goals: Expedite customer time-to-value and ROI, by driving business-aligned innovation and solutions assurance within Pivotal’s Data Fabric technologies.

Drive customer adoption and autonomy across the full spectrum of Pivotal Data technologies through best-in-class data science and data engineering services, with a deep emphasis on knowledge transfer.


Procedural Languages


MPP Architectural Overview

Think of it as multiple PostGreSQL servers

Workers

Master


User-Defined Functions (UDFs) �  PostgreSQL/Greenplum provide lots of flexibility in defining your own functions.

�  Simple UDFs are SQL queries with calling arguments and return types.

Definition: CREATE FUNCTION times2(INT) RETURNS INT AS $$ SELECT 2 * $1 $$ LANGUAGE sql;

Execution: SELECT times2(1); times2 -‐-‐-‐-‐-‐-‐-‐-‐ 2 (1 row)


�  The interpreter/VM of the language ‘X’ is installed on each node of the Greenplum Database Cluster

•  Data Parallelism: -  PL/X piggybacks on

Greenplum’s MPP architecture

•  Allows users to write Greenplum/PostgreSQL functions in the R/Python/Java, Perl, pgsql or C languages Standby

Master

…

Master Host

SQL

Interconnect

Segment Host Segment Segment




PL/X : X in {pgsql, R, Python, Java, Perl, C etc.}


Data Parallelism

� Little or no effort is required to break up the problem into a number of parallel tasks, and there exists no dependency (or communication) between those parallel tasks.

� Examples: –  Measure the height of each student in a classroom (explicitly

parallelizable by student) –  MapReduce –  Individual machine learning models per customer/region/etc.


Intro to PL/Python �  Procedural languages need to be installed on each database used.

�  Syntax is like normal Python function inside a SQL wrapper

�  Need to specify return type

CREATE FUNCTION pymax (a integer, b integer) RETURNS integer AS $$ if a > b: return a return b $$ LANGUAGE plpythonu;

SQL wrapper

SQL wrapper

Normal Python


Returning Results �  Postgres primitive types (int, bigint, text, float8, double precision, date, NULL etc.) �  Composite types can be returned by creating a composite type in the database:

CREATE TYPE named_value AS ( name text, value integer );

�  Then you can return a list, tuple or dict (not sets) which reference the same structure as the table: CREATE FUNCTION make_pair (name text, value integer) RETURNS named_value AS $$ return [ name, value ] # or alternatively, as tuple: return ( name, value ) # or as dict: return { "name": name, "value": value } # or as an object with attributes .name and .value $$ LANGUAGE plpythonu;

�  For functions which return multiple rows, prefix “setof” before the return type


Parallelisation

Separate data into units using GROUP BY

Aggregate input for function using array_agg

SELECT name , plp.linregr(array_agg(x), array_agg(y))

FROM plp.test_data GROUP BY name ORDER BY name;


Benefits of Procedural Languages

� Easy to bring your code to the data.

� When SQL falls short leverage your Python/R/Java/C experience quickly.

� Apply Python/R packages across terabytes of data with minimal overhead or additional requirements.

� Results are already in the database system, ready for further analysis or storage.


Text Analytics for Churn Prediction Customer

A major telecom company

Business Problem

Reducing churn through more accurate models

Challenges

�  Existing models only used structured features

�  Call center memos had poor structure and had lots of typos

Solution

�  Built sentiment analysis models to predict churn and topic models to understand topics of conversation in call center memos

�  Achieved 16% improvement in ROC curve for Churn Prediction


Predicting Customs Clearance Intercepts Customer

A major courier delivery services company

Business Problem

7% of shipments into the US are intercepted by Customs based on package description.

Challenge

�  Data Volume: Previously not able to build models with rich features for billions of packages.

�  Data Variety: Customer’s prior technology did not allow them to extract features from unstructured shipment documents.

Solution

•  Processed shipment description documents using PL/Python. Tokenized, stemmed, removed stop words, etc.

•  Built features to summarize the shippers’ prior shipment and interception history

•  Built a model which predicts the propensity of a package being intercepted (AUC =0.86)

Intercepted Not Intercepted


Network User Behavior Anomaly Detection Customer

A major global financial service enterprise

Business Problem

Detect anomalous user behaviour in the global enterprise computer network

Challenges

10 Billions events in 6 months; 15K+ network devices; No existing SIEM solutions can model user behavioral resource access baseline and enable anomaly detection in an adaptive and scalable architecture.

Solution

An innovative Graph Mining based algorithmic framework with advanced Machine Learning. Network topology and temporal behaviors are both modeled. (Patent pending). Implemented in MPP and PL/R, enabling parallel model training and behavior risk scoring. Successfully identified DLP violating anomalous users.


MADlib


MADlib: The Origin

•  First mention of MAD analytics was at VLDB 2009 MAD Skills: New Analysis Practices for Big Data J. Hellerstein, J. Cohen, B. Dolan, M. Dunlap, C. Welton (with help from: Noelle Sio, David Hubbard, James Marca) http://db.cs.berkeley.edu/papers/vldb09-madskills.pdf

•  MADlib project initiated in late 2010:

Greenplum Analytics team and Prof. Joe Hellerstein

UrbanDictionary mad (adj.): an adjective used to enhance a noun. 1- dude, you got skills. 2- dude, you got mad skills

•  Open Source!https://github.com/madlib/madlib

•  Works on Greenplum DB, PostgreSQL and also HAWQ & Impala

•  Active development by Pivotal -  Latest Release: v1.5 (Apr 2014)

•  Downloads and Docs: http://madlib.net/


MADlib In-Database Functions

Predictive Modeling Library

Linear Systems •  Sparse and Dense Solvers

Matrix Factorization •  Single Value Decomposition (SVD) •  Low-Rank

Generalized Linear Models •  Linear Regression •  Logistic Regression •  Multinomial Logistic Regression •  Cox Proportional Hazards •  Regression •  Elastic Net Regularization •  Sandwich Estimators (Huber white,

clustered, marginal effects)

Machine Learning Algorithms •  Principal Component Analysis (PCA) •  Association Rules (Affinity Analysis, Market

Basket) •  Topic Modeling (Parallel LDA) •  Decision Trees •  Ensemble Learners (Random Forests) •  Support Vector Machines •  Conditional Random Field (CRF) •  Clustering (K-means) •  Cross Validation

Descriptive Statistics

Sketch-based Estimators •  CountMin (Cormode-

Muthukrishnan) •  FM (Flajolet-Martin) •  MFV (Most Frequent

Values) Correlation Summary

Support Modules

Array Operations Sparse Vectors Random Sampling Probability Functions


Get in touch

Feel free to contact me about PL/Python & PL/R, or more generally about Data Science and opportunities available.

@ianhuston

ihuston @ gopivotal.com http://www.ianhuston.net

BUILT FOR THE SPEED OF BUSINESS

Additional Line 18 Point Verdana

data science amsterdam - massively parallel processing with procedural languages

Technology