About big data

From: http://www.slideshare.net/timoelliott/ukisug2014-big-data-presentation

Types of big data:

• Human generated data:
• Swipe of credit card
• Scan of bar code
• Actions captured on mobile phones
• Machine generated data:
• Data logers
• Sensors

From hindsight, to insight to foresight:

• Descriptive: what did happen
• Diagnostic: why did it happen
• Predictive: will it happen again
• Prescriptive: how can we make it happen

Types of data (sorted by % use in descending order) - ref2:

• Time series
• Business transactions
• Geospatial/location
• Graph (network)
• Clickstream
• Health records
• Sensor
• Image
• Genomic

According to Gartner (Ref3):

• Rising up to the peak of inflated expectation:
• Data science
• Predictive analytics
• In the downslope of desillusion
• Complexe event processing
• Big Data
• Content analytics

Big data can be used for:

• Engage and empower data consumers:
• Discovering new business opportunities
• Identify new product opportunities
• More reliable decision making
• Plan and optimize: Improve operation efficiency
• Personalize experience

Data lake: is a storage repository that holds a vast amount of raw data in its native format until it is needed.
HTAP = OLTP + OLAP = new generation of in-memory data platforms that can perform both online transaction processing (OLTP) and online analytical processing (OLAP) without requiring data duplication.

Ref 2: from http://www.paradigm4.com/wp-content/uploads/2014/06/P4-data-scientist-survey-FINAL.pdf

Ref 3: Gartner 2014 hype cycle: http://www.gartner.com/newsroom/id/2819918

Hadoop - Beyond Map Reduce, Pig and Hive

Hadoop 2.0

6/8/14 - Hadoop Maturity Summit: http://gigaom.com/2014/06/08/hadoop-maturity-summits/

Yarn

Part of Hadoop 2.0 Yarn (Yet Another Resource Manager) is a resource manager that enable non MapReduce jobs to work on Hadoop and leverage HDFS. YARN provide a generic resource management framework for implementing distributed applications.
MapReduce only allows batch processing, but YARN unlock the real time data processing on Hadoop.

Spark - http://projects.apache.org/projects/spark.html

Apache Spark is a fast and general engine for large-scale data processing. It offers high-level APIs in Java, Scala and Python as well as a rich set of libraries including stream processing, machine learning, and graph analytics.
Spark powers a stack of high-level tools including Shark for SQL, MLib for machine learning , GraphX and Spark Streaming.

MLib

MLib is spark's scallable machine learning library. It fits into Spar's APIs and interoperates with NumPy in Python. It can leverage any Hadoop data source (HDFS, HBase or local files).
Machine Learning Library (MLib) guide: http://spark.apache.org/docs/latest/mllib-guide.html

Cassandra - http://cassandra.apache.org/

Cassandra is a massively scalable open source NoSQL. Cassandra is perfect for managing large amounts of structured, semi-structured, and unstructured data across multiple data centers and the cloud. Cassandra delivers continuous availability, linear scalability, and operational simplicity across many commodity servers with no single point of failure, along with a powerful dynamic data model designed for maximum flexibility and fast response times.
Cassandra sports a “masterless” architecture meaning all nodes are the same.

Kafka - http://kafka.apache.org/

Kafka is a distributed pub/sub and message queuing system.
Kafka consumes streams called topics that are partitioned and replicated across multiple machine named brokers.
A single Kafka broker can handle hundreds of megabytes of reads and writes per second from thousands of clients. Kafka is designed to allow a single cluster to serve as the central data backbone for a large organization. It can be elastically and transparently expanded without downtime. Data streams are partitioned and spread over a cluster of machines to allow data streams larger than the capability of any single machine and to allow clusters of co-ordinated consumers. Kafka has a modern cluster-centric design that offers strong durability and fault-tolerance guarantees. Messages are persisted on disk and replicated within the cluster to prevent data loss. Each broker can handle terabytes of messages without performance impact.
Basic messaging terminology:

• Kafka maintains feeds of message in categories called topics.
• Process that publish message to a Kafka topic is called producer.
• Process that subscribes to topics and process the feed of a published message is called consumer.
• Kafka is run in a cluster comprised of one or more servers each of which is called a broker.

Storm
http://vimeo.com/40972420

HBase
http://hbase.apache.org/book.html#other.info.videos
Use Apache HBase™ when you need random, realtime read/write access to your Big Data. This project's goal is the hosting of very large tables -- billions of rows X millions of columns -- atop clusters of commodity hardware. Apache HBase is an open-source, distributed, versioned, non-relational database modeled after Google's Bigtable. Apache HBase provides Bigtable-like capabilities on top of Hadoop and HDFS.

HDFS is a distributed file system that is well suited for the storage of large files. Its documentation states that it is not, however, a general purpose file system, and does not provide fast individual record lookups in files. HBase, on the other hand, is built on top of HDFS and provides fast record lookups (and updates) for large tables.

Both Cassandra and HBase are NoSQL databases, a term for which you can find numerous definitions. Generally, it means you cannot manipulate the database with SQL. However, Cassandra has implemented CQL (Cassandra Query Language), the syntax of which is obviously modeled after SQL. Cassandra uses the Gossip protocol for internode communications, and Gossip services are integrated with the Cassandra software. HBase relies on Zookeeper -- an entirely separate distributed application -- to handle corresponding tasks.

Hadoop - Intro

http://youtu.be/d2xeNpfzsYI

Map Reduce process: 1)Map (key:value pair)-- 2)Shuffle-sort (sort, computation phase)-- 3)Reduce (aggregate results)

Toolcase

Front end: HTML/CSS/JS, Wireframing

Backend: SSJS, DB, Framework, Data Pipelines

APIs: Client-side templating, HTTP, SOA/REST/JSON

URI Uniform Resource Identifier

Is made of the URL (Uniform Resource Locator) and URN (Uniform Resource Name)

Syntax: URI consists of a URI scheme name (such as "HTTP","ftp","mailto","crid","file") followed by a colon character, then a scheme specific part. More at http://en.wikipedia.org/wiki/URI_scheme#Generic_syntax

http://en.wikipedia.org/wiki/URI#Examples_of_URI_references

"http" specifies the 'scheme' name, "en.wikipedia.org" is the 'authority', "/wiki/URI" the 'path' pointing to this article, and "#Examples_of_URI_references" is a 'fragment' pointing to this section.

 REST API:

• the base URI for the web API, such as http://example.com/resources/
• the Internet media type of the data supported by the web API. This is often JSON but can be any other valid Internet media type provided that it is a valid hypertext standard.
• the set of operations supported by the web API using HTTP methods (e.g., GET, PUT, POST, or DELETE).
• The API must be hypertext driven.[16]

RESTful web API HTTP methods

Resource	GET	PUT	POST	DELETE
Collection URI, such as http://example.com/resources	List the URIs and perhaps other details of the collection's members.	Replace the entire collection with another collection.	Create a new entry in the collection. The new entry's URI is assigned automatically and is usually returned by the operation.	Delete the entire collection.
Element URI, such as http://example.com/resources/item17	Retrieve a representation of the addressed member of the collection, expressed in an appropriate Internet media type.	Replace the addressed member of the collection, or if it doesn't exist, createit.	Not generally used. Treat the addressed member as a collection in its own right andcreate a new entry in it.	Delete the addressed member of the collection.

productService.GetProduct("1")

http://someurl/product/1

REST(ful) API is an API that follows the REST architectural style.

REST stands for Representational State Transfer,. Its HTTP version uses the four HTTP methods GET, POST, PUT and DELETE to execute different operations.

A RESTful web service is a web API implemented using HTTP and REST principles. It is a collection of resources, with four defined aspects:

The following table shows how the HTTP methods are typically used to implement a web API.

Where SOAP tries to model the exchange between client and server as calls to objects, REST tries to be faithful to the web domain. So when calling a web service written in SOAP, you may write

in REST, you may call a url with HTTP GET:

New BI Comers

New ETL player:
Lavastorm Analytics:
Provide tools to analyst to extract, transform, clean and manipulate data to deliver analysis.
The strength is that there is no need for programing, use visual based application to assemble blocks representing transformation rules (like modern ETL tools) with output be caned reports/dashboards, alerts, case management module with embedded logic.

New DB:
NeoTechnology:
New type of database technology that marries NoSQL database (i.e. direct interface with Java and online world, schema less) with VLDB and its data storage and management capabilities.

New light BI:
Jaspersoft:

LogiXML:

Hadoop tools

Karmasphere

Abstraction layer and tool on top of Hadoop that enables to query Hadoop files using SQL. The tool is comprised of interactive web interface where users can create a project that will contain a set of queries. These queries can be shared and parametrised. The output can as well be used by tool like Tableau.
A drawback is that it doesn't solve the latency issue of Hadoop bu rather provide a tracker that shows which queries are running and which one are finished.

HBase

Open source, non-relational, NoSQL database in Java that seats on top of Hadoop and thus taking advantage of HDFS to store BigTables. It uses the programing concept of key-value representing data with "row"/"column-family"/"column"/"timsetamp"/"value"


Flume

Mahout

Hive

Pig

A/B Testing

What is A/B testing? It is conducting experiment to optimize customer experience. If the test is a success, we will then use the winning test to influence the behavior of the customer to our benefit (the one defined in the test).

An A/B has 4 steps:
1) Analyse the data
2) Form an hypothesis
3) Construct an experiment
4) Interpret the result

1) Analyse the data

Quantitative data tells where to test (get this data by behavior analysis...).
Qualitative data tells what to test (get this data via survey, feedback...).

2) Form an hypothesis

"If [variable], then [result] because [rational]."
[variable] = the element that is modified:

• Call to action wording, size, color and placement
• Headline of product description
• Form length and types of fields
• Layout and style of website
• Product pricing and promotional offers
• Images on landing and product pages
• Amount of text on the page (short vs. long)

[result] = predicted outcome derived from the analysis of data is 1) (more sign-up, improved CTA, % increase, number goal etc...)
[rational] = what will be proven wrong if the test fails?

ex:
"If the navigation band is removed from the check-out page, then the conversion rate at each step will increase because our website analytics shows that traffic drops due to click on these links."

3) Experimentation

Every test have 3 parts:

• Content: what does it says?
• Design: how does it look?
• Tech: how does it work?

4) Interpret the result

We want to answer the question: How confident are we that the results are not due to chance?

Mathematically, the conversion rate is represented by a binomial random variable, which is a fancy way of saying that it can have two possible values: conversion or non-conversion.

(The binomial distribution with parameters n and p is the discrete probability distribution of the number of successes in a sequence of n independent yes/no experiments, each of which yields success with probability p.)

To declare a winning we want the result of the test to be statistically significant at 95% which mean that only 5% probability that the result was due to chance.

This equate to having all the tested Conversion rate within 1.96 SE from the average conversion rate C.

To avoid doing repeated experiments, statistics has a neat trick in its toolbox. There is a concept called standard error, which tells how much deviation from average conversion rate (C) can be expected if this experiment is repeated multiple times. Smaller the deviation, more confident you can be about estimating true conversion rate.

For a given conversion rate (C) and number of trials (n), standard error is calculated as:

Standard Error (SE) = Square root of (C * (1-C) / n)

you can be sure with 95% confidence that your true conversion rate lies within this range: C % ± 2 * SE
(1.96 to be exact).

Testing the statistical significance corresponds to


p is the conversion rate that we are observing.

GOING INTO THE NITY DETAILS -- NEED CLEANING AND CLARIFICATION
http://www.slideshare.net/dj4b1n/ab-testing-pitfalls-and-lessons-learned-at-spotify-31935130

A/B Testing Pitfalls and Lessons Learned at Spotify from dj4b1n

The likelihood of obtaining a certain value under a given distribution is measured by its p-value 

If there is a low likelihood that a change is due to chance alone, we call our results statistically significant

Statistical significance is measure by alpha which typical has a value of 5% or 1%.
Alternatively P(significant) = 0.05 or 0.01
There is a conversion from alpha to z-Score.
The Z-score tells us how far a particular value is from the mean (and what the corresponding likelihood is)

Standard deviation (σ) tells us how spread out the numbers are

Test of statistical significance:

Null hypothesis is that the conversion rate of the control is no less than the conversion rate of the experiment:

H0: CR - CRc =<0

CRc is the conversion rate of the control.

Alternative hypothesis is that the experimental page has a higher conversion rate. This is what we want to test.

Conversion rate are normally distributed (conversion = success).

What we are looking for is if the difference between the two conversion rate is large enough to conclude that the treatment altered the behavior.

Random variable for the test will be: X = CR - CRc

Null hypothesis: H0: p-Value =<0

We use the Z-score to derive the p-value:

Z-Score = (CRc - Cre)/√(σc^2+σe^2)

p-Value = NORMDIST(Z-Score, 0, 1, TRUE)

if p-Value >95% then we can reject Null hypothesis and the A/B test is good to go!