Data Science with Spark - Training at SparkSummit (East)

Data Science Training
Spark
“I want to die on Mars but not on impact”
— Elon Musk, interview with Chris Anderson
“The shrewd guess, the fertile hypothesis, the courageous leap to a
tentative conclusion – these are the most valuable coin of the thinker at
work” -- Jerome Seymour Bruner
"There are no facts, only interpretations." - Friedrich Nietzsche
"If you torture the data long enough, it will confess to anything." – Hal Varian,
Computer Mediated Transactions!
------ We are not going to hang data by it’s legs !!
http://training.databricks.com/workshop/datasci.pdf

ADVANCED: DATA SCIENCE WITH APACHE SPARK
Data Science applications with Apache Spark combine the scalability of Spark and the
distributed machine learning algorithms.
This material expands on the “Intro to Apache Spark” workshop. Lessons focus on
industry use cases for machine learning at scale, coding examples based on public
data sets, and leveraging cloud-based notebooks within a team context. Includes
limited free accounts on Databricks Cloud.
Topics covered include:
Data transformation techniques based on both Spark SQL and functional
programming in Scala and Python.
Predictive analytics based on MLlib, clustering with KMeans, building classiﬁers
with a variety of algorithms and text analytics – all with emphasis on an
iterative cycle of feature engineering, modeling, evaluation.
Visualization techniques (matplotlib, ggplot2, D3, etc.) to surface insights.
Understand how the primitives like Matrix Factorization are implemented in a
distributed parallel framework from the designers of MLlib
Several hands-on exercises using datasets such as Movielens, Titanic, State Of
the Union speeches, and RecSys Challenge 2015.
Prerequisites:
Intro to Apache Spark workshop or
equivalent (e.g., Spark Developer Certiﬁcate)
Experience coding in Scala, Python, SQL
Have some familiarity with Data Science
topics (e.g., business use cases)

Agenda
•  Detailed)agenda)In)Google)doc)
•  https://docs.google.com/document/d/
1T9AkXUmL6gDYTpAEEgqsy9hfJtqGlavjnGzMqzUwDf
E/edit)

Goals
•  Patterns):)Data)wrangling)(Transform,)Model)&)Reason))with)Spark)
o  Use)RDDs,)Transformations)and)Actions)in)the)context)of)a)Data)Science)Problem,)an)Algorithms))&)a)
Dataset)
•  Spend)time)working)through)MLlib))
•  Balance)between)internals)&)handsWon)
o  Internals)from)Reza,)the)MLlib)lead)
•  ~65%)of)time)on)Databricks)Cloud)&)Notebooks)
o  Take)the)time)to)get)familiar)with)the)Interface)&)the)Data)Science)Cloud)
o  Make*mistakes,*experiment,…*
•  Good)Time)for)this)course,)this)version)
o  Will)miss)many)of)the)gory)details)as)the)framework)evolves)
•  Summarized)materials)for)a)3)day)course)
o  Even)if)we)don’t)ﬁnish)the)exercises)today,)that)is)ﬁne)
o  Complete)the)work)at)home)W)There*are*also**homework*notebooks*
o  Ask)us)questions)@ksankar,@pacoid,*@reza_zadeh,*@mhfalaki,*@andykonwinski,*@xmeng,*
@michaelarmbrust,*@tathadas*

Tutorial)Outline:)
morning' a)ernoon'
o  Welcome)+)Ge3ng)Started)(Krishna))
o  Databricks)Cloud)mechanics)(Andy)))
o  Ex)3):)Clustering)B)In)which)we)explore)SegmenFng)
Frequent)InterGallacFcHoppers)(Krishna))
o  Ex)0:)PreBFlight)Check)(Krishna)) o  Ex)4):)RecommendaFon)(Krishna))
o  DataScience)DevOps)B)IntroducFon)to)Spark)(Krishna)) o  Theory):)Matrix)FactorizaFon,)SVD,…)(Reza))
o  Ex)1:)MLlib):)StaFsFcs,)Linear)Regression)(Krishna)) o  OnBline)kBmeans,)spark)streaming)(Reza))
o  MLlib)Deep)Dive)–)Lecture)(Reza))
o  Design)Philosophy,)APIs)
o  Ex)2:)In)which)we)explore)Disasters,)Trees,)
ClassiﬁcaFon)&)the)Kaggle)CompeFFon)(Krishna))
o  Random)Forest,)Bagging,)Data)DeBcorrelaFon)
o  Ex)5):)Mood)of)the)UnionBText)AnalyFcs(Krishna))
o  In)which)we)analyze)the)Mood)of)the)naFon)from)
inferences)on)SOTU)by)the)POTUS)(State)of)the)Union)
Addresses)by)The)President)Of)the)US))
o  Deepdive)B)Leverage)parallelism)of)RDDs,)sparse)
vectors,)etc)(Reza))
o  Ex)99):)RecSys)2015)Challenge)(Krishna))
o  Ask)Us)Anything)B)Panel)

Introducing:)
Reza Zadeh
@Reza_Zadeh
Hossein Falaki
@mhfalaki
Andy Konwinski
@andykonwinski
Krishna Sankar
@ksankar
Paco Nathan
@pacoid
Xiangrui Meng
@xmeng
Michael Armbrust
@michaelarmbrust
Tathagata Das
@tathadas

About Me
o Chief Data Scientist at BlackArrow.tv
o Have been speaking at OSCON, PyCon, Pydata,
Strata et al
o Reviewer “Machine Learning with Spark”
o Picked up co-authorship Second Edition of “Fast
Data Processing with Spark”
o Have done lots of things:
• 
• 
• 
• 
• 
• 
o  @ksankar, doubleclix.wordpress.com ksankar42@gmail.com
The)Nuthead)band)!)

Pre-requisites
①  Register)&)Download)data)from)Kaggle.)
We)cannot)distribute)Kaggle)data.)
Moreover)you)need)an)account)to)submit)entries)
a)  Setup)an)account)in)Kaggle)(www.kaggle.com))
b)  We)will)be)using)the)data)from)the)competition)“Titanic:)
Machine)Learning)from)Disaster”)
c)  Download)data)from)
http://www.kaggle.com/c/titanicWgettingStarted)
②  Register)for)RecSys)2015)Competition)
a)  http://2015.recsyschallenge.com/)

Welcome +
Getting Started
9:00

Everyone will receive a username/password for one !
of the Databricks Cloud shards. Use your laptop and browser to login there.
We ﬁnd that cloud-based notebooks are a simple way to get started using
Apache Spark – as the motto “Making Big Data Simple” states.
Please create and run a variety of notebooks on your account throughout the
tutorial. These accounts will remain open long enough for you to export your
work.
See the product page or FAQ for more details, or contact Databricks to register
for a trial account.
10
Getting Started: Step 1

11
Getting Started: Step 1 – Credentials
url: https://class01.cloud.databricks.com/
user: student-777
pass: 93ac11xq23z5150
cluster: student-777

12
Open in a browser window, then click on the navigation menu in the top/left
corner:

13
The next columns to the right show folders,!
and scroll down to click on databricks_guide

14
Scroll to open the 01 Quick Start notebook, then follow the discussion
about using key features:

15
See /databricks-guide/01 Quick Start !
Key Features:
•  Workspace / Folder / Notebook
•  Code Cells, run/edit/move/comment
• Markdown
•  Results
•  Import/Export

16
Click on the Workspace menu and create your !
own folder (pick a name):

17
Navigate to /_DataScience 
Hover on its drop-down menu, on the right side:

18
Navigate to /_DataScience 
Hover on its drop-down menu, on the right side:
Click Clone:

19
Then create a clone of this folder in the folder that you just created:

20
Now let’s get started with the coding exercise!
We’ll deﬁne an initial Spark app in three lines of code:
Click on _00.pre-ﬂight-check
Getting Started: Coding Exercise

21
Attach your cluster – same as your username:

Data Science :
The art of building a model with known knowns, which when let loose,
works with unknown unknowns!
http://smartorg.com/2013/07/valuepoint19/
The
World

Knowns&
&
&
&
&
&
Unknowns&
You
UnKnown& Known&
o  Others)know,)you)don’t) o  What)we)do)
o  Facts,)outcomes)or)
scenarios)we)have)not)
encountered,)nor)
considered)
o  “Black)swans”,)outliers,)
long)tails)of)probability)
distribuFons)
o  Lack)of)experience,)
imaginaFon)
o  PotenFal)facts,)
outcomes)we)
are)aware,)but)
not))with)
certainty)
o  StochasFc)
processes,)
ProbabiliFes)
o 
o 
o 
o 
o 
o

The curious case of the Data Scientist
o Data Scientist is multi-faceted & Contextual
o Data Scientist should be building Data
Products
o Data Scientist should tell a story
http://doubleclix.wordpress.com/2014/01/25/theWcuriousWcaseWofWtheWdataWscientistWprofession/

Data Science - Context
o  Scalable)Model)
Deployment)
o  Big)Data)
automation)&)
purpose)built)
appliances)(soft/
hard))
o  Manage)SLAs)&)
response)times)
o  Volume)
o  Velocity)
o  Streaming)Data)
o  Canonical)form)
o  Data)catalog)
o  Data)Fabric)across)the)
organization)
o  Access)to)multiple)
sources)of)data))
o  Think)Hybrid)–)Big)
Data)Apps,)Appliances)
&)Infrastructure)
Collect Store Transform
o  Metadata)
o  Monitor)counters)&)
Metrics)
o  Structured)vs.)MultiW
structured)
o  Flexible)&)Selectable)
!  Data)Subsets))
!  Attribute)sets)
o  Reﬁne)model)with)
!  Extended)Data)
subsets)
!  Engineered)
Attribute)sets)
o  Validation)run)across)a)
larger)data)set)
Reason Model Deploy
Data Management
Data Science
o  Dynamic)Data)Sets)
o  2)way)keyWvalue)tagging)of)
datasets)
o  Extended)attribute)sets)
o  Advanced)Analytics)
ExploreVisualize Recommend Predict
o  Performance)
o  Scalability)
o  Refresh)Latency)
o  InWmemory)Analytics)
o  Advanced)Visualization)
o  Interactive)Dashboards)
o  Map)Overlay)
o  Infographics)
"  Bytes to Business
a.k.a. Build the full
stack
"  Find Relevant Data
For Business
"  Connect the Dots

Volume
Velocity
Variety
Data Science - Context
Context
Connect
edness
Intelligence
Interface
Inference
o  Three Amigos
o  Interface = Cognition
o  Intelligence = Compute(CPU) & Computational(GPU)
o  Infer Significance & Causality

Day in the life of a (super) Model
Intelligence
Inference
Data Representation
Interface
Algorithms)
Parameters)Agributes)
Data)(Scoring))
Model)
SelecFon)
Reason)&)
Learn)
Models)
Visualize,)
Recommend,)
Explore)
Model)Assessment)
Feature)SelecFon)Dimensionality)ReducFon)

The'Sense'&'Sensibility'of'a'
DataScien7st'DevOps'
Factory)=)OperaFonal)
Lab)=)InvesFgaFve)
http://doubleclix.wordpress.com/2014/05/11/theWsenseW
sensibilityWofWaWdataWscientistWdevops/)

RDD – The workhorse of Spark
o Resilient Distributed Datasets
• 
o Transformations – create RDDs
• 
o Actions – Get values
• 
o We will apply these operations during this tutorial

MLlib Hands-on
Stats, Linear Regression
9:20

Algorithm spectrum
o 
o 
o 
o 
o 
o 
o 
o 
)
o 
o 
o 
o 
o 
o 
o 
)
Machine0
Learning0
Cute0Math0
Ar6ﬁcial0
Intelligence0

Session – 1 : MLlib - Statistics &
Linear Regression
1.  Notebook):)01_StatsLRW1)
①  Read)car)data)
②  Stats)(Guided))
③  Correlation)(Guided))
④  Coding)ExerciseW21WTemplate)
(Correlation))
①  CEW21WSolution)
3.  Linear)Regression)
①  LR)(Guided))
②  CEW22WTemplate((LR)on)Car)
Data))
①  CEW22WSolution)
②  Explain)

Linear Regression - API
LabeledPoint The)features)and)labels)of)a)data)point
LinearModel weights,)intercept
LinearRegressionModelBase predict()
LinearRegressionModel
LinearRegressionWithSGD train(cls,)data,)iteraFons=100,)step=1.0,)miniBatchFracFon=1.0,)
iniFalWeights=None,)regParam=1.0,)regType=None,)intercept=False)
LassoModel LeastBsquares)ﬁt)with)an)L1)penalty)term.
LassoWithSGD train(cls,)data,)iteraFons=100,)step=1.0,)regParam=1.0,)
miniBatchFracFon=1.0,iniFalWeights=None)
RidgeRegressionModel LeastBsquares)ﬁt)with)an)L2)penalty)term.
RidgeRegressionWithSGD train(cls,)data,)iteraFons=100,)step=1.0,)regParam=1.0,)miniBatchFracFon=1.0,)
iniFalWeights=None)

MLlib - Deep Dive
•  Design Philosophy & APIs
•  Algorithms - Regression, SGD et al
•  Interfaces
9:50

Reza Zadeh
Distributed Machine Learning"
on Spark
@Reza_Zadeh | http://reza-zadeh.com

Outline
Data ﬂow vs. traditional network programming
Spark computing engine
Optimization Examples
Matrix Computations
MLlib + {Streaming, GraphX, SQL}
Future of MLlib

Data Flow Models
Restrict the programming interface so that the
system can do more automatically
Express jobs as graphs of high-level operators
» System picks how to split each operator into tasks
and where to run each task
» Run parts twice fault recovery
Biggest example: MapReduce
Map
Map
Map
Reduce
Reduce

Spark Computing Engine
Extends a programming language with a
distributed collection data-structure
» “Resilient distributed datasets” (RDD)
Open source at Apache
» Most active community in big data, with 50+
companies contributing
Clean APIs in Java, Scala, Python
Community: SparkR, soon to be merged

Key Idea
Resilient Distributed Datasets (RDDs)
» Collections of objects across a cluster with user
controlled partitioning & storage (memory, disk, ...)
» Built via parallel transformations (map, ﬁlter, …)
» The world only lets you make make RDDs such that
they can be:
Automatically rebuilt on failure

MLlib History
MLlib is a Spark subproject providing machine
learning primitives

Initial contribution from AMPLab, UC Berkeley

Shipped with Spark since Sept 2013

MLlib: Available algorithms
classification: logistic regression, linear SVM,"
naïve Bayes, least squares, classification tree
regression: generalized linear models (GLMs),
regression tree
collaborative filtering: alternating least squares (ALS),
non-negative matrix factorization (NMF)
clustering: k-means||
decomposition: SVD, PCA
optimization: stochastic gradient descent, L-BFGS

Optimization
At least two large classes of optimization
problems humans can solve:"

»  Convex
»  Spectral

Optimization Example: Convex Optimization

Logistic Regression
data$=$spark.textFile(...).map(readPoint).cache()$
$
w$=$numpy.random.rand(D)$
$
for$i$in$range(iterations):$
$$$$gradient$=$data.map(lambda$p:$
$$$$$$$$(1$/$(1$+$exp(Bp.y$*$w.dot(p.x))))$*$p.y$*$p.x$
$$$$).reduce(lambda$a,$b:$a$+$b)$
$$$$w$B=$gradient$
$
print$“Final$w:$%s”$%$w$

Separable Updates
Can be generalized for
»  Unconstrained optimization
»  Smooth or non-smooth
»  LBFGS, Conjugate Gradient, Accelerated
Gradient methods, …

Logistic Regression Results
0
500
1000
1500
2000
2500
3000
3500
4000
1
5
10
20
30
RunningTime(s)
Number of Iterations
Hadoop
Spark
110 s / iteration
ﬁrst iteration 80 s
further iterations 1 s
100 GB of data on 50 m1.xlarge EC2 machines
!

Behavior with Less RAM
68.8
58.1
40.7
29.7
11.5
0
20
40
60
80
100
0%
25%
50%
75%
100%
Iterationtime(s)
% of working set in memory

Optimization Example: Spectral Program

Spark PageRank
Given directed graph, compute node
importance. Two RDDs:
»  Neighbors (a sparse graph/matrix)
»  Current guess (a vector)
"
Using cache(), keep neighbor list in RAM

Spark PageRank
Using cache(), keep neighbor lists in RAM
Using partitioning, avoid repeated hashing
Neighbors
(id, edges)
Ranks
(id, rank)
join
partitionBy
join
join
…

PageRank Results
171
72
23
0
50
100
150
200
Timeperiteration(s)
Hadoop
Basic Spark
Spark + Controlled
Partitioning

Spark PageRank
Generalizes!to!Matrix!Multiplication,!opening!many!algorithms!
from!Numerical!Linear!Algebra!

Distributing Matrix Computations

Distributing Matrices
How to distribute a matrix across machines?
»  By Entries (CoordinateMatrix)
»  By Rows (RowMatrix)
»  By Blocks (BlockMatrix)
All of Linear Algebra to be rebuilt using these
partitioning schemes
As!of!version!1.3!

Distributing Matrices
Even the simplest operations require thinking
about communication e.g. multiplication

How many different matrix multiplies needed?
»  At least one per pair of {Coordinate, Row,
Block, LocalDense, LocalSparse} = 10
»  More because multiplies not commutative

MLlib - Hands On #2 – Kaggle
Competition Predicting Titanic
Survivors:
•  Feature Engineering
•  Classification Algorithms(Random forest)
•  Submission & Leaderboard scores
10:45

Data Science “folk knowledge” (1 of A)
o "If you torture the data long enough, it will confess to anything." – Hal Varian,
Computer Mediated Transactions
o Learning = Representation + Evaluation + Optimization
o It’s Generalization that counts
• 
o Data alone is not enough
• 
o Machine Learning is not magic – one cannot get something from nothing
• 
• 
A few useful things to know about machine learning - by Pedro Domingos
http://dl.acm.org/citation.cfm?id=2347755

Classification - Spark API
•  Logistic)Regression)
•  SVMWithSGD)
•  DecisionTrees)
•  Data)as)LabelledPoint)(we)will)see)in)a)moment))
•  DecisionTree.trainClassiﬁer(data,)numClasses,)categoricalFeaturesInfo,)impurity="gini",)
maxDepth=4,)maxBins=100))
•  Impurity)–)“entropy”)or)“gini”)
•  maxBins)=)control)to)throttle)communication)at)the)expense)of)accuracy)
o  Larger)=)Higher)Accuracy)
o  Smaller)=)less)communication)(as)#)of)bins)=)number)of)instances))
•  Data)Adaptive)–)i.e.)decision)tree)samples)on)the)driver)and)ﬁgures)out)the)bin)spacing)i.e.)the)
places)you)slice)for)binning)
•  Spark*=*Intelligent*Framework*D*need*this*for*scale*

Lookout for these interesting Spark
features
•  Concept)of)Labeled)Point)&)how)to)create)an)
RDD)of)LPs)
•  Print)the)tree)
•  Calculate)Accuracy)&)MSE)from)RDDs)

Anatomy Of a Kaggle
Competition

Kaggle Data Science Competitions
o Hosts Data Science Competitions
o Competition Attributes:
• 
• 
• 
• 
• 
• 
• 
• 
• 
•

Titanic)Passenger)Metadata)
•  Small)
•  3)Predictors)
•  Class)
•  Sex)
•  Age)
•  Survived?)
http://www.ohgizmo.com/2007/03/21/romain-jerome-titanic
http://flyhigh-by-learnonline.blogspot.com/2009/12/at-movies-sherlock-holmes-2009.html
City Bike Sharing Prediction
(Washington DC)
Walmart Store Forecasting

Train.csv)
Taken)from)Titanic)
Passenger)Manifest)
Variable' Descrip7on'
Survived) 0BNo,)1=yes)
Pclass) Passenger)Class)()1st,2nd,3rd)))
Sibsp) Number)of)Siblings/Spouses)Aboard)
Parch) Number)of)Parents/Children)Aboard)
Embarked) Port)of)EmbarkaFon)
o  C)=)Cherbourg)
Titanic)Passenger)Metadata)
•  Small)
•  3)Predictors)
•  Class)
•  Sex)
•  Age)
•  Survived?)

Test.csv)
Submission
o 418 lines; 1st column should have 0 or 1 in each line
o Evaluation:
•

Data Science “folk knowledge” (Wisdom of Kaggle)
Jeremy’s Axioms
o Iteratively explore data
o Tools
• 
o Get your head around data
• 
o Don’t over-complicate
o If people give you data, don’t assume that you
need to use all of it
o Look at pictures !
o History of your submissions – keep a tab
o Don’t be afraid to submit simple solutions
• 
Ref: http://blog.kaggle.com/2011/03/23/getting-in-shape-for-the-sport-of-data-sciencetalk-by-
jeremy-howard/

Session-2 : Kaggle, Classification &
Trees
1.  Notebook):)04_TitanicW01)
①  Read)Training)Data)
②  Henry)the)Sixth)Model)
③  Submit)to)Kaggle)
①  Decision)Tree)Model)
②  CEW31)Template)
①  Create)Randomforest)Model)
②  Predict)Testset)
③  Submit)to)Kaggle)
①  CEW32)Solution)
①  RandomForest)Model)
②  Predict)Testset)
③  Submit)Solution)2)
②  Discussion)about)Models)

Trees, Forests & Classification
•  Discuss)Random)Forest)
o Boosting,)Bagging)
o Data)deWcorrelation))
•  Why)it)didn’t)do)better)in)Titanic)dataset)
•  Data)Science)Folk)Wisdom)
o http://www.slideshare.net/ksankar/dataWscienceWfolkW
knowledge)

Results
Algorithm' Kaggle'Score'
Gender)based)model) ~0.7655)Rank):)1276)
Decision)Tree) ~0.775124,)Rank):)1147)
Random)Forest)) ~0.77512)Gender)Based)Model)
Why didn’t RF do better ? Bias/Variance
See next slide
Henry0the0Sixth,0Part02,0Act04,0Scene020

Why didn’t RF do better ? Bias/Variance
o High Bias
• 
• 
• 
• 
• 
o High Variance
• 
• 
• 
• 
Prediction Error!
Training !
Error!
Ref: Strata 2013 Tutorial by Olivier Grisel
Learning
Curve
Need*more*features*or*more*
complex*model*to*improve*
Need*more*data*
to*improve*
'Bias is a learner’s tendency to consistently learn the same wrong thing.' -- Pedro Domingos
hgp://www.slideshare.net/ksankar/dataBscienceBfolkBknowledge)

Decision Tree – Best Practices
maxDepth) Tune)with)Data/Model)SelecFon)
maxBins) Set)low,)monitor)communicaFons,)increase)if)needed)
#)RDD)parFFons) Set)to)#)of)cores)
•  Usually the recommendation is that the RDD partitions should be over
partitioned ie “more partitions than cores”, because tasks take different
times, we need to utilize the compute power and in the end they average
out
•  But for Machine Learning especially trees, all tasks are approx equal
computationally intensive, so over partitioning doesn’t help
•  Joe Bradley talk (reference below) has interesting insights
hgps://speakerdeck.com/jkbradley/mllibBdecisionBtreesBatBsfBscalaBbamlBmeetup)
DecisionTree.trainClassiﬁer(data,)numClasses,)categoricalFeaturesInfo,)
impurity="gini",)maxDepth=4,)maxBins=100))

◦  “Output)of)weak)classifiers)into)a)powerful)
commigee”)
◦  Final)PredicFon)=)weighted)majority)vote))
◦  Later)classifiers)get)misclassified)points))
!  With)higher)weight,))
!  So)they)are)forced))
!  To)concentrate)on)them)
◦  AdaBoost)(AdapFveBoosting))
◦  BoosFng)vs)Bagging)
!  Bagging)–)independent)trees)<B)Spark)shines)here)
!  BoosFng)–)successively)weighted)
Boosting
"  Goal!
◦  Model Complexity (-)!
◦  Variance (-)!
◦  Prediction Accuracy (+)!

◦  Builds)large)collecFon)of)deBcorrelated)trees)&)averages)them)
◦  Improves)Bagging)by)selecFng)i.i.d*)random)variables)for)
spli3ng)
◦  Simpler)to)train)&)tune)
◦  “Do0remarkably0well,0with0very0liIle0tuning0required”)–)ESLII)
◦  Less)suscepFble)to)over)ﬁ3ng)(than)boosFng))
◦  Many)RF)implementaFons)
!  Original)version)B)FortranB77)!)By)Breiman/Cutler)
!  Python,)R,)Mahout,)Weka,)Milk)(ML)toolkit)for)py),)matlab))
!  And)of)course,)Spark)!)
* i.i.d – independent identically distributed
+ http://www.stat.berkeley.edu/~breiman/RandomForests/cc_home.htm
Random Forests+
"  Goal!

Random Forests
•  While)Boosting)splits)based)on)best)among)all)variables,)RF)splits)based)
on)best)among)randomly*chosen)variables)
•  )Simpler)because)it)requires)two)variables)–)no.)of)Predictors)(typically)√k))
&)no.)of)trees)(500)for)large)dataset,)150)for)smaller))
•  Error)prediction)
o  For)each)iteration,)predict)for)dataset)that)is)not)in)the)sample)(OOB)data))
o  Aggregate)OOB)predictions)
o  Calculate)Prediction)Error)for)the)aggregate,)which)is)basically)the)OOB)estimate)of)error)rate)
•  Can)use)this)to)search)for)optimal)#)of)predictors)
o  We)will)see)how)close)this)is)to)the)actual)error)in)the)Heritage)Health)Prize)
•  Assumes)equal)cost)for)misWprediction.)Can)add)a)cost)function)
•  Proximity)matrix)&)applications)like)adding)missing)data,)dropping)
outliers))
Ref: R News Vol 2/3, Dec 2002
Statistical Learning from a Regression Perspective : Berk
A Brief Overview of RF by Dan Steinberg

◦  Two)Step)
!  Develop)a)set)of)learners)
!  Combine)the)results)to)develop)a)composite)predictor)
◦  Ensemble)methods)can)take)the)form)of:)
!  Using)different)algorithms,))
!  Using)the)same)algorithm)with)different)se3ngs)
!  Assigning)different)parts)of)the)dataset)to)different)
classifiers)
◦  Bagging)&)Random)Forests)are)examples)of)
ensemble)method))
Ref: Machine Learning In Action
Ensemble Methods
"  Goal!

Deepdive : Leverage
parallelism of RDDs, sparse
vectors, etc.
11:30

Clustering - Hands On :
•  Normalization & Centering
•  Clustering
•  Optimizing k based on cohesively of the
clusters (WSSE)
1:30

Data Science “folk knowledge” (3 of A)
o More Data Beats a Cleverer Algorithm
• 
• 
o Learn many models, not Just One
• 
• 
• 
o Simplicity Does not necessarily imply Accuracy
o Representable Does not imply Learnable
• 
o Correlation Does not imply Causation o  http://doubleclix.wordpress.com/2014/03/07/a-glimpse-of-google-nasa-peter-norvig/
o  A few useful things to know about machine learning - by Pedro Domingos
!  http://dl.acm.org/citation.cfm?id=2347755

Session-3 : Clustering
1.  Notebook):)07_ClusterW1)
①  Read)Data)
②  Cluster)
③  Modeling)ExerciseW41WTemplate)
①  MEW41WSolution)
②  Center)and)Scale)
③  Cluster)
④  Inspect)centroid)
⑤  CEW42WTemplate):)Cluster)
Semantics)
②  Cluster)Semantics)W)Discussion)

Clustering - Theory
•  Clustering)is)unsupervised)learning)
•  While)the)computers)can)dissect)a)dataset)into)“similar”)clusters,)it)
still)needs)human)direction)&)domain)knowledge)to)interpret)&)
guide)
•  Two)types:)
o  Centroid)based)clustering)–)kWmeans)clustering)
o  Tree)based)Clustering)–)hierarchical)clustering)
•  Spark)implements)the)Scalable)Kmeans++))
o  Paper):)http://theory.stanford.edu/~sergei/papers/vldb12Wkmpar.pdf)

Lookout for these interesting Spark
features
•  Application)of)Statistics)toolbox)
•  Center)&)Scale)RDD)
•  Filter)RDDs)

Clustering - API
•  from)pyspark.mllib.clustering)import)KMeans)
•  Kmeans.train)
•  train(cls,)data,)k,)maxIterations=100,)runs=1,)initializationMode="kW
means||"))
•  K)=)number)of)clusters)to)create,)default=2)
•  initializationMode)=)The)initialization)algorithm.)This)can)be)either)
"random")to)choose)random)points)as)initial)cluster)centers,)or)"kW
means||")to)use)a)parallel)variant)of)kWmeans++)(Bahmani)et)al.,)Scalable)
KWMeans++,)VLDB)2012).)Default:)kWmeans||)
•  KMeansModel.predict)
•  Maps)a)point)to)a)cluster)

Interpretation
C#' AVG' Interpreta7on'
1)
2) )
)
Not)that)acFve)–)Give)them)flight)coupons)to)encourage)them)to)fly)more.)Ask)them)why)they)are)not)
flying.)May)be)they)are)flying)to)desFnaFons)(say)Jupiter))where)InterGallacFc)has)less)gates)
3)
4)
5)
Best0Customers,0S6ll0lots0
of0nonLflying0miles0
Very0ac6ve0onLline.0Why0are0they0coming0to0us0instead0of0Amazon0?0
Note0:00
•  This0is0just0a0sample0interpreta6on.0
•  In0real0life0we0would0“noodle”0over0the0clusters0&0tweak0
them0to0be0useful,0interpretable0and0dis6nguishable.0
•  May0be030is0more0suited0to0create0targeted0promo6ons0

Epilogue
•  KMeans)in)Spark)has)enough)controls)
•  It)does)a)decent)job)
•  We)were)able)to)control)the)clusters)based)on)our)experience)(2)
cluster)is)too)low,)10)is)too)high,)5)seems)to)be)right))
•  We)can)see)that)the)Scalable)KMeans)has)control)over)runs,)
parallelism)et)al.)(Home)work):)Explore)the)scalability))
•  We)were)able)to)interpret)the)results)with)domain)knowledge)and)
arrive)at)a)scheme)to)solve)the)business)opportunity)
•  Naturally)we)would)tweak)the)clusters)to)ﬁt)the)business)viability.)
20)clusters)with)corresponding)promotion)schemes)are)unwieldy,)
even)if)the)WSSE)is)the)minimum.)

Recommendation - Hands On :
•  Movie Lens - Medium Data
•  Movie Lens - Large Data (Homework)
2:00

hgps://doubleclix.wordpress.com/2014/03/07/aBglimpseBofBgoogleBnasaBpeterBnorvig/)

Session-4 : Recommendation at
Scale
1.  NotebookW10_RecoW1)
①  Read)Movielens)medium)
data)
②  CEW51)Template)–)Partition)
Data)
2.  NotebookW11_RecoW2)
②  ALS)Slide)
③  Train)ALS)&)Predict)
④  Calculate)Model)
Performance)

Recommendation & Personalization - Spark
Automated*AnalyticsW)Let)Data)tell)story)
Feature)Learning,)AI,)Deep)Learning)
Learning*Models*W)ﬁt)parameters)
as)it)gets)more)data))
Dynamic*Models*–)model)
selection)based)on)context)
o  Knowledge)Based)
o  Demographic)Based)
o  Content)Based)
o  Collaborative)Filtering)
o  Item)Based)
o  User)Based)
o  Latent)Factor)based)
o  User)Rating)
o  Purchased)
o  Looked/Not)purchased)
Spark)(in)1.1.0))implements)the)user)based)ALS)collaboraFve)ﬁltering)
Ref:))
ALS)B)CollaboraFve)Filtering)for)Implicit)Feedback)Datasets,)Yifan)Hu);)AT&T)Labs.,)Florham)Park,)NJ);)Koren,)Y.);)Volinsky,)C.)
ALSBWR)B)LargeBScale)Parallel)CollaboraFve)Filtering)for)the)Nezlix)Prize,)Yunhong)Zhou,)Dennis)Wilkinson,)Robert)Schreiber,)Rong)P

Spark Collaborative Filtering API
•  ALS.train(cls,)ratings,)rank,)iterations=5,)
lambda_=0.01,)blocks=W1))
•  ALS.trainImplicit(cls,)ratings,)rank,)iterations=5,)
lambda_=0.01,)blocks=W1,)alpha=0.01))
•  MatrixFactorizationModel.predict(self,)user,)
product))
•  MatrixFactorizationModel.predictAll(self,)
usersProducts))

Theory : Matrix Factorization, SVD,…
On-line k-means, spark streaming
2:30

Singular Value Decomposition on Spark

Singular Value Decomposition
Two cases
»  Tall and Skinny
»  Short and Fat (not really)
»  Roughly Square
SVD method on RowMatrix takes care of
which one to call.

Tall and Skinny SVD
Gets%us%%%V%and%the%
singular%values%
Gets%us%%%U%by%one%
matrix%multiplication%

Square SVD
ARPACK: Very mature Fortran77 package for
computing eigenvalue decompositions"

JNI interface available via netlib-java"

Distributed using Spark – how?

Square SVD via ARPACK
Only interfaces with distributed matrix via
matrix-vector multiplies

The result of matrix-vector multiply is small.
The multiplication can be distributed.

Square SVD
With 68 executors and 8GB memory in each,
looking for the top 5 singular vectors

Communication-Efﬁcient
All pairs similarity on Spark (DIMSUM)

All pairs Similarity
All pairs of cosine scores between n vectors
»  Don’t want to brute force (n choose 2) m
»  Essentially computes
"
Compute via DIMSUM
»  Dimension Independent Similarity
Computation using MapReduce

Intuition
Sample columns that have many non-zeros
with lower probability. "

On the ﬂip side, columns that have fewer non-
zeros are sampled with higher probability."

Results provably correct and independent of
larger dimension, m.

MLlib + {Streaming, GraphX, SQL}

A General Platform
Spark Core
Spark
Streaming"
real-time
Spark SQL
structured
GraphX
graph
MLlib
machine
learning
…
Standard libraries included with Spark

Beneﬁt for Users
Same engine performs data extraction, model
training and interactive queries

…
DFS
read
DFS
write
parse
DFS
read
DFS
write
train
DFS
read
DFS
write
query
DFS
DFS
read
parse
train
query
Separate engines
Spark

MLlib + Streaming
As of Spark 1.1, you can train linear models in
a streaming fashion, k-means as of 1.2

Model weights are updated via SGD, thus
amenable to streaming

More work needed for decision trees

MLlib + SQL
df = context.sql(“select latitude, longitude from tweets”)!
model = pipeline.fit(df)!
DataFrames in Spark 1.3! (March 2015)
Powerful coupled with new pipeline API

Goals for next version
Tighter integration with DataFrame and spark.ml API

Accelerated gradient methods & Optimization interface

Model export: PMML (current export exists in Spark 1.3, but
not PMML, which lacks distributed models)

Scaling: Model scaling (e.g. via Parameter Servers)

Research Goal: General
Distributed Optimization
Distribute%CVX%by%
backing%CVXPY%with%
PySpark%
%
EasyBtoBexpress%
distributable%convex%
programs%
%
Need%to%know%less%
math%to%optimize%
complicated%
objectives%

Most active open source community in big data
200+ developers, 50+ companies contributing
Spark Community
0
50
100
150
Contributors in past year

Continuing Growth
source: ohloh.net
Contributors per month to Spark

Spark and ML
Spark has all its roots in research, so we hope
to keep incorporating new ideas!

Hands On :
•  Mood Of the Union
•  RecSys 2015 Challenge
3:15

The Art of ELO Ranking
& Super Bowl XLIX
o The real formula is
o Not what is written on the glass !
o But then that is Hollywood !
Ref : Who is #1, Princeton University Presshttps://doubleclix.wordpress.com/2015/01/20/theWartWofWnﬂWrankingWtheWeloWalgorithmWandWﬁvethirtyeight/)

Session-5 : Mood Of the Union –
Data Science on SOTU by POTUS
1.  DataScience/12_SOTUW1)
①  Read)BO)
②  CEW61)Template):)Has)BO)
changed)since)2014)?)
2.  DataScience/13_SOTUW2)
②  Read)GW)
③  Preprocess)
④  CEW62)Template):)What)
mood)the)country)was)in)
1790W1796)vs.)2009W2015)
3.  Notebook):)14_SOTUW3)
②  Homework)
①  GWB)vs)Clinton)
②  WJC)vs)AL)
③  Discussions ) ))

WJC)
Mood Of The Nation
JFK)
FDR)
GW)
AL)
JFK)

Epilogue
•  Interesting)Exercise)
•  Highlights)
o  Map-reduce in a couple of lines !
•  But it is not exactly the same as Hadoop Mapreduce (see the excellent blog by Sean Owen1)
o  Set differences using substractByKey
o  Ability to sort a map by values (or any arbitrary function, for that matter)
•  To)Explore)as)homework:)
o  TF-IDF in
http://spark.apache.org/docs/latest/mllib-feature-extraction.html#tf-idf
hgp://blog.cloudera.com/blog/2014/09/howBtoBtranslateBfromBmapreduceBtoBapacheBspark/)

Session-7 : Predict Buying Pattern
Recsys 2015 Challenge
1.  Notebook):)
99_RecsysW2015W1)
1.  Read)Data)
2.  Explore)Options)
3.  CEW71)

Notebooks to Explore
•  Run)thru)all)the)
homework)in)you)
Databricks)cloud)

82
GraphX:
spark.apache.org/docs/latest/graphx-programming-guide.html
Key Points:
•  graph-parallel systems
•  importance of workﬂows
•  optimizations

83
PowerGraph: Distributed Graph-Parallel Computation on Natural Graphs!
J. Gonzalez, Y. Low, H. Gu, D. Bickson, C. Guestrin!
graphlab.org/ﬁles/osdi2012-gonzalez-low-gu-bickson-guestrin.pdf
Pregel: Large-scale graph computing at Google!
Grzegorz Czajkowski, et al.!
googleresearch.blogspot.com/2009/06/large-scale-graph-computing-at-google.html
GraphX: Uniﬁed Graph Analytics on Spark!
Ankur Dave, Databricks!
databricks-training.s3.amazonaws.com/slides/graphx@sparksummit_2014-07.pdf
Advanced Exercises: GraphX!
databricks-training.s3.amazonaws.com/graph-analytics-with-graphx.html
GraphX: Further Reading…

// http://spark.apache.org/docs/latest/graphx-programming-guide.html
import org.apache.spark.graphx._
import org.apache.spark.rdd.RDD
case class Peep(name: String, age: Int)
val nodeArray = Array(
(1L, Peep("Kim", 23)), (2L, Peep("Pat", 31)),
(3L, Peep("Chris", 52)), (4L, Peep("Kelly", 39)),
(5L, Peep("Leslie", 45))
)
val edgeArray = Array( 84
GraphX: Example – simple traversals

85
GraphX: Example – routing problems
What is the cost to reach node 0 from any other node in the graph? This is a
common use case for graph algorithms, e.g., Djikstra

86
Run /<your folder>/_DataSCience/08.graphx!
in your folder:
GraphX: Coding Exercise

Case Studies: Apache Spark, DBC, etc.
Additional details about production deployments for Apache Spark can be
found at:
https://cwiki.apache.org/conﬂuence/display/SPARK/Powered+By
+Spark
https://databricks.com/blog/category/company/partners
http://go.databricks.com/customer-case-studies
88

Case Studies: Automatic Labs
89
Spark Plugs Into Your Car!
Rob Ferguson!
spark-summit.org/east/2015/talk/spark-plugs-into-
your-car
ﬁnance.yahoo.com/news/automatic-labs-turns-
databricks-cloud-140000785.html
Automatic creates personalized driving habit dashboards
•  wanted to use Spark while minimizing investment in DevOps
•  provides data access to non-technical analysts via SQL
•  replaced Redshift and disparate ML tools with single platform
•  leveraged built-in visualization capabilities in notebooks to generate
dashboards easily and quickly

Spark at Twitter: Evaluation & Lessons Learnt!
Sriram Krishnan!
slideshare.net/krishflix/seattle-spark-meetup-spark-at-twitter
•  Spark can be more interactive, efficient than MR
•  support for iterative algorithms and caching
•  more generic than traditional MapReduce
•  Why is Spark faster than Hadoop MapReduce?
•  fewer I/O synchronization barriers
•  less expensive shuffle
•  the more complex the DAG, the greater the !
performance improvement
90
Case Studies: Twitter

Pearson uses Spark Streaming for next generation
adaptive learning platform
Dibyendu Bhattacharya
databricks.com/blog/2014/12/08/pearson-uses-
spark-streaming-for-next-generation-adaptive-
learning-platform.html
91
• Kafka + Spark + Cassandra + Blur, on AWS on a YARN cluster
• single platform/common API was a key reason to replace Storm
with Spark Streaming
• custom Kafka Consumer for Spark Streaming, using Low Level
Kafka Consumer APIs
• handles: Kafka node failures, receiver failures, leader changes,
committed offset in ZK, tunable data rate throughput
Case Studies: Pearson

Unlocking Your Hadoop Data with Apache Spark and CDH5
Denny Lee
slideshare.net/Concur/unlocking-your-hadoop-data-with-apache-
spark-and-cdh5
92
• leading provider of spend management solutions and services
• delivers recommendations based on business users’ travel and
expenses – “to help deliver the perfect trip”
• use of traditional BI tools with Spark SQL allowed analysts to
make sense of the data without becoming programmers
• needed the ability to transition quickly between Machine Learning
(MLLib), Graph (GraphX), and SQL usage
• needed to deliver recommendations in real-time
Case Studies: Concur

Stratio Streaming: a new approach to Spark Streaming
David Morales, Oscar Mendez
spark-summit.org/2014/talk/stratio-streaming-a-new-
approach-to-spark-streaming
93
• Stratio Streaming is the union of a real-time messaging bus with a
complex event processing engine atop Spark Streaming
• allows the creation of streams and queries on the ﬂy
• paired with Siddhi CEP engine and Apache Kafka
• added global features to the engine such as auditing and statistics
Case Studies: Stratio

Collaborative Filtering with Spark!
Chris Johnson!
slideshare.net/MrChrisJohnson/collaborative-filtering-with-spark
•  collab filter (ALS) for music recommendation
•  Hadoop suffers from I/O overhead
•  show a progression of code rewrites, converting a !
Hadoop-based app into efficient use of Spark
94
Case Studies: Spotify

Guavus Embeds Apache Spark !
into its Operational Intelligence Platform !
Deployed at the World’s Largest Telcos
Eric Carr
databricks.com/blog/2014/09/25/guavus-embeds-apache-spark-into-its-operational-intelligence-
platform-deployed-at-the-worlds-largest-telcos.html
95
•  4 of 5 top mobile network operators, 3 of 5 top Internet
backbone providers, 80% MSOs in NorAm
•  analyzing 50% of US mobile data traffic, +2.5 PB/day
•  latency is critical for resolving operational issues before they
cascade: 2.5 MM transactions per second
•  “analyze first” not “store first ask questions later”
Case Studies: Guavus

Case Studies: Radius Intelligence
96
From Hadoop to Spark in 4 months, Lessons Learned!
Alexis Roos!
http://youtu.be/o3-lokUFqvA
•  building a full SMB index took 12+ hours using !
Hadoop and Cascading
•  pipeline was difﬁcult to modify/enhance
•  Spark increased pipeline performance 10x
•  interactive shell and notebooks enabled data scientists !
to experiment and develop code faster
•  PMs and business development staff can use SQL to !
query large data sets

community:
spark.apache.org/community.html
events worldwide: goo.gl/2YqJZK
video+preso archives: spark-summit.org
resources: databricks.com/spark-training-resources
workshops: databricks.com/spark-training

Data Science with Spark - Training at SparkSummit (East)

Data Science with Spark - Training at SparkSummit (East)

More Related Content

What's hot

Viewers also liked

Similar to Data Science with Spark - Training at SparkSummit (East)

More from Krishna Sankar

Recently uploaded

Data Science with Spark - Training at SparkSummit (East)