This the multi-page printable view of this section. Click here to print.

Return to the regular view of this page.

Applications

Applications

We will discuss each of these applications in more detail# Applications

1 - Big Data Use Cases Survey

Big Data Use Cases Survey

This section covers 51 values of X and an overall study of Big data that emerged from a NIST (National Institute for Standards and Technology) study of Big data. The section covers the NIST Big Data Public Working Group (NBD-PWG) Process and summarizes the work of five subgroups: Definitions and Taxonomies Subgroup, Reference Architecture Subgroup, Security and Privacy Subgroup, Technology Roadmap Subgroup and the Requirements andUse Case Subgroup. 51 use cases collected in this process are briefly discussed with a classification of the source of parallelism and the high and low level computational structure. We describe the key features of this classification.

NIST Big Data Public Working Group

This unit covers the NIST Big Data Public Working Group (NBD-PWG) Process and summarizes the work of five subgroups: Definitions and Taxonomies Subgroup, Reference Architecture Subgroup, Security and Privacy Subgroup, Technology Roadmap Subgroup and the Requirements and Use Case Subgroup. The work of latter is continued in next two units.

Presentation Overview (45)

Introduction to NIST Big Data Public Working

The focus of the (NBD-PWG) is to form a community of interest from industry, academia, and government, with the goal of developing a consensus definitions, taxonomies, secure reference architectures, and technology roadmap. The aim is to create vendor-neutral, technology and infrastructure agnostic deliverables to enable big data stakeholders to pick-and-choose best analytics tools for their processing and visualization requirements on the most suitable computing platforms and clusters while allowing value-added from big data service providers and flow of data between the stakeholders in a cohesive and secure manner.

Video Introduction (13:02)

Definitions and Taxonomies Subgroup

The focus is to gain a better understanding of the principles of Big Data. It is important to develop a consensus-based common language and vocabulary terms used in Big Data across stakeholders from industry, academia, and government. In addition, it is also critical to identify essential actors with roles and responsibility, and subdivide them into components and sub-components on how they interact/ relate with each other according to their similarities and differences.

For Definitions: Compile terms used from all stakeholders regarding the meaning of Big Data from various standard bodies, domain applications, and diversified operational environments. For Taxonomies: Identify key actors with their roles and responsibilities from all stakeholders, categorize them into components and subcomponents based on their similarities and differences. In particular data Science and Big Data terms are discussed.

Video Taxonomies (7:42)

Reference Architecture Subgroup

The focus is to form a community of interest from industry, academia, and government, with the goal of developing a consensus-based approach to orchestrate vendor-neutral, technology and infrastructure agnostic for analytics tools and computing environments. The goal is to enable Big Data stakeholders to pick-and-choose technology-agnostic analytics tools for processing and visualization in any computing platform and cluster while allowing value-added from Big Data service providers and the flow of the data between the stakeholders in a cohesive and secure manner. Results include a reference architecture with well defined components and linkage as well as several exemplars.

Video Architecture (10:05)

Security and Privacy Subgroup

The focus is to form a community of interest from industry, academia, and government, with the goal of developing a consensus secure reference architecture to handle security and privacy issues across all stakeholders. This includes gaining an understanding of what standards are available or under development, as well as identifies which key organizations are working on these standards. The Top Ten Big Data Security and Privacy Challenges from the CSA (Cloud Security Alliance) BDWG are studied. Specialized use cases include Retail/Marketing, Modern Day Consumerism, Nielsen Homescan, Web Traffic Analysis, Healthcare, Health Information Exchange, Genetic Privacy, Pharma Clinical Trial Data Sharing, Cyber-security, Government, Military and Education.

Video Security (9:51)

Technology Roadmap Subgroup

The focus is to form a community of interest from industry, academia, and government, with the goal of developing a consensus vision with recommendations on how Big Data should move forward by performing a good gap analysis through the materials gathered from all other NBD subgroups. This includes setting standardization and adoption priorities through an understanding of what standards are available or under development as part of the recommendations. Tasks are gather input from NBD subgroups and study the taxonomies for the actors' roles and responsibility, use cases and requirements, and secure reference architecture; gain understanding of what standards are available or under development for Big Data; perform a thorough gap analysis and document the findings; identify what possible barriers may delay or prevent adoption of Big Data; and document vision and recommendations.

Video Technology (4:14)

Interfaces Subgroup

This subgroup is working on the following document: NIST Big Data Interoperability Framework: Volume 8, Reference Architecture Interface.

This document summarizes interfaces that are instrumental for the interaction with Clouds, Containers, and HPC systems to manage virtual clusters to support the NIST Big Data Reference Architecture (NBDRA). The Representational State Transfer (REST) paradigm is used to define these interfaces allowing easy integration and adoption by a wide variety of frameworks. . This volume, Volume 8, uses the work performed by the NBD-PWG to identify objects instrumental for the NIST Big Data Reference Architecture (NBDRA) which is introduced in the NBDIF: Volume 6, Reference Architecture.

This presentation was given at the 2nd NIST Big Data Public Working Group (NBD-PWG) Workshop in Washington DC in June 2017. It explains our thoughts on deriving automatically a reference architecture form the Reference Architecture Interface specifications directly from the document.

The workshop Web page is located at

The agenda of the workshop is as follows:

The Web cas of the presentation is given bellow, while you need to fast forward to a particular time

You are welcome to view other presentations if you are interested.

Requirements and Use Case Subgroup

The focus is to form a community of interest from industry, academia, and government, with the goal of developing a consensus list of Big Data requirements across all stakeholders. This includes gathering and understanding various use cases from diversified application domains.Tasks are gather use case input from all stakeholders; derive Big Data requirements from each use case; analyze/prioritize a list of challenging general requirements that may delay or prevent adoption of Big Data deployment; develop a set of general patterns capturing the essence of use cases (not done yet) and work with Reference Architecture to validate requirements and reference architecture by explicitly implementing some patterns based on use cases. The progress of gathering use cases (discussed in next two units) and requirements systemization are discussed.

Video Requirements (27:28)

51 Big Data Use Cases

This units consists of one or more slides for each of the 51 use cases - typically additional (more than one) slides are associated with pictures. Each of the use cases is identified with source of parallelism and the high and low level computational structure. As each new classification topic is introduced we briefly discuss it but full discussion of topics is given in following unit.

Presentation 51 Use Cases (100)

Government Use Cases

This covers Census 2010 and 2000 - Title 13 Big Data; National Archives and Records Administration Accession NARA, Search, Retrieve, Preservation; Statistical Survey Response Improvement (Adaptive Design) and Non-Traditional Data in Statistical Survey Response Improvement (Adaptive Design).

Video Government Use Cases (17:43)

Commercial Use Cases

This covers Cloud Eco-System, for Financial Industries (Banking, Securities & Investments, Insurance) transacting business within the United States; Mendeley - An International Network of Research; Netflix Movie Service; Web Search; IaaS (Infrastructure as a Service) Big Data Business Continuity & Disaster Recovery (BC/DR) Within A Cloud Eco-System; Cargo Shipping; Materials Data for Manufacturing and Simulation driven Materials Genomics.

Video Commercial Use Cases (17:43)

Defense Use Cases

This covers Large Scale Geospatial Analysis and Visualization; Object identification and tracking from Wide Area Large Format Imagery (WALF) Imagery or Full Motion Video (FMV) - Persistent Surveillance and Intelligence Data Processing and Analysis.

Video Defense Use Cases (15:43)

Healthcare and Life Science Use Cases

This covers Electronic Medical Record (EMR) Data; Pathology Imaging/digital pathology; Computational Bioimaging; Genomic Measurements; Comparative analysis for metagenomes and genomes; Individualized Diabetes Management; Statistical Relational Artificial Intelligence for Health Care; World Population Scale Epidemiological Study; Social Contagion Modeling for Planning, Public Health and Disaster Management and Biodiversity and LifeWatch.

Video Healthcare and Life Science Use Cases (30:11)

Deep Learning and Social Networks Use Cases

This covers Large-scale Deep Learning; Organizing large-scale, unstructured collections of consumer photos; Truthy: Information diffusion research from Twitter Data; Crowd Sourcing in the Humanities as Source for Bigand Dynamic Data; CINET: Cyberinfrastructure for Network (Graph) Science and Analytics and NIST Information Access Division analytic technology performance measurement, evaluations, and standards.

Video Deep Learning and Social Networks Use Cases (14:19)

Research Ecosystem Use Cases

DataNet Federation Consortium DFC; The ‘Discinnet process’, metadata -big data global experiment; Semantic Graph-search on Scientific Chemical and Text-based Data and Light source beamlines.

Video Research Ecosystem Use Cases (9:09)

Astronomy and Physics Use Cases

This covers Catalina Real-Time Transient Survey (CRTS): a digital, panoramic, synoptic sky survey; DOE Extreme Data from Cosmological Sky Survey and Simulations; Large Survey Data for Cosmology; Particle Physics: Analysis of LHC Large Hadron Collider Data: Discovery of Higgs particle and Belle II High Energy Physics Experiment.

Video Astronomy and Physics Use Cases (17:33)

Environment, Earth and Polar Science Use Cases

EISCAT 3D incoherent scatter radar system; ENVRI, Common Operations of Environmental Research Infrastructure; Radar Data Analysis for CReSIS Remote Sensing of Ice Sheets; UAVSAR Data Processing, DataProduct Delivery, and Data Services; NASA LARC/GSFC iRODS Federation Testbed; MERRA Analytic Services MERRA/AS; Atmospheric Turbulence - Event Discovery and Predictive Analytics; Climate Studies using the Community Earth System Model at DOE’s NERSC center; DOE-BER Subsurface Biogeochemistry Scientific Focus Area and DOE-BER AmeriFlux and FLUXNET Networks.

Video Environment, Earth and Polar Science Use Cases (25:29)

Energy Use Case

This covers Consumption forecasting in Smart Grids.

Video Energy Use Case (4:01)

Features of 51 Big Data Use Cases

This unit discusses the categories used to classify the 51 use-cases. These categories include concepts used for parallelism and low and high level computational structure. The first lesson is an introduction to all categories and the further lessons give details of particular categories.

Presentation Features (43)

Summary of Use Case Classification

This discusses concepts used for parallelism and low and high level computational structure. Parallelism can be over People (users or subjects), Decision makers; Items such as Images, EMR, Sequences; observations, contents of online store; Sensors – Internet of Things; Events; (Complex) Nodes in a Graph; Simple nodes as in a learning network; Tweets, Blogs, Documents, Web Pages etc.; Files or data to be backed up, moved or assigned metadata; Particles/cells/mesh points. Low level computational types include PP (Pleasingly Parallel); MR (MapReduce); MRStat; MRIter (Iterative MapReduce); Graph; Fusion; MC (Monte Carlo) and Streaming. High level computational types include Classification; S/Q (Search and Query); Index; CF (Collaborative Filtering); ML (Machine Learning); EGO (Large Scale Optimizations); EM (Expectation maximization); GIS; HPC; Agents. Patterns include Classic Database; NoSQL; Basic processing of data as in backup or metadata; GIS; Host of Sensors processed on demand; Pleasingly parallel processing; HPC assimilated with observational data; Agent-based models; Multi-modal data fusion or Knowledge Management; Crowd Sourcing.

Video Summary of Use Case Classification (23:39)

Database(SQL) Use Case Classification

This discusses classic (SQL) database approach to data handling with Search&Query and Index features. Comparisons are made to NoSQL approaches.

Video Database (SQL) Use Case Classification (11:13)

NoSQL Use Case Classification

This discusses NoSQL (compared in previous lesson) with HDFS, Hadoop and Hbase. The Apache Big data stack is introduced and further details of comparison with SQL.

Video NoSQL Use Case Classification (11:20)

Other Use Case Classifications

This discusses a subset of use case features: GIS, Sensors. the support of data analysis and fusion by streaming data between filters.

Video Use Case Classifications I (12:42) This discusses a subset of use case features: Pleasingly parallel, MRStat, Data Assimilation, Crowd sourcing, Agents, data fusion and agents, EGO and security.

Video Use Case Classifications II (20:18)

This discusses a subset of use case features: Classification, Monte Carlo, Streaming, PP, MR, MRStat, MRIter and HPC(MPI), global and local analytics (machine learning), parallel computing, Expectation Maximization, graphs and Collaborative Filtering.

Video Use Case Classifications III (17:25)

\TODO{These resources have not all been checked to see if they still exist this is currently in progress}

Resources

Some of the links bellow may be outdated. Please let us know the new links and notify us of the outdated links.

2 - Cloud Computing

Cloud Computing

We describe the central role of Parallel computing in Clouds and Big Data which is decomposed into lots of Little data running in individual cores. Many examples are given and it is stressed that issues in parallel computing are seen in day to day life for communication, synchronization, load balancing and decomposition. Cyberinfrastructure for e-moreorlessanything or moreorlessanything-Informatics and the basics of cloud computing are introduced. This includes virtualization and the important as a Service components and we go through several different definitions of cloud computing.

Gartner’s Technology Landscape includes hype cycle and priority matrix and covers clouds and Big Data. Two simple examples of the value of clouds for enterprise applications are given with a review of different views as to nature of Cloud Computing. This IaaS (Infrastructure as a Service) discussion is followed by PaaS and SaaS (Platform and Software as a Service). Features in Grid and cloud computing and data are treated. We summarize the 21 layers and almost 300 software packages in the HPC-ABDS Software Stack explaining how they are used.

Cloud (Data Center) Architectures with physical setup, Green Computing issues and software models are discussed followed by the Cloud Industry stakeholders with a 2014 Gartner analysis of Cloud computing providers. This is followed by applications on the cloud including data intensive problems, comparison with high performance computing, science clouds and the Internet of Things. Remarks on Security, Fault Tolerance and Synchronicity issues in cloud follow. We describe the way users and data interact with a cloud system. The Big Data Processing from an application perspective with commercial examples including eBay concludes section after a discussion of data system architectures.

Parallel Computing (Outdated)

We describe the central role of Parallel computing in Clouds and Big Data which is decomposed into lots of ‘‘Little data’’ running in individual cores. Many examples are given and it is stressed that issues in parallel computing are seen in day to day life for communication, synchronization, load balancing and decomposition.

Presentation Parallel Computing (33)

Decomposition

We describe why parallel computing is essential with Big Data and distinguishes parallelism over users to that over the data in problem. The general ideas behind data decomposition are given followed by a few often whimsical examples dreamed up 30 years ago in the early heady days of parallel computing. These include scientific simulations, defense outside missile attack and computer chess. The basic problem of parallel computing – efficient coordination of separate tasks processing different data parts – is described with MPI and MapReduce as two approaches. The challenges of data decomposition in irregular problems is noted.

Parallel Computing in Society

This lesson from the past notes that one can view society as an approach to parallel linkage of people. The largest example given is that of the construction of a long wall such as that (Hadrian’s wall) between England and Scotland. Different approaches to parallelism are given with formulae for the speed up and efficiency. The concepts of grain size (size of problem tackled by an individual processor) and coordination overhead are exemplified. This example also illustrates Amdahl’s law and the relation between data and processor topology. The lesson concludes with other examples from nature including collections of neurons (the brain) and ants.

Parallel Processing for Hadrian’s Wall

This lesson returns to Hadrian’s wall and uses it to illustrate advanced issues in parallel computing. First We describe the basic SPMD – Single Program Multiple Data – model. Then irregular but homogeneous and heterogeneous problems are discussed. Static and dynamic load balancing is needed. Inner parallelism (as in vector instruction or the multiple fingers of masons) and outer parallelism (typical data parallelism) are demonstrated. Parallel I/O for Hadrian’s wall is followed by a slide summarizing this quaint comparison between Big data parallelism and the construction of a large wall.

Resources

Introduction

We discuss Cyberinfrastructure for e-moreorlessanything or moreorlessanything-Informatics and the basics of cloud computing. This includes virtualization and the important ‘as a Service’ components and we go through several different definitions of cloud computing.Gartner’s Technology Landscape includes hype cycle and priority matrix and covers clouds and Big Data. The unit concludes with two simple examples of the value of clouds for enterprise applications. Gartner also has specific predictions for cloud computing growth areas.

Presentation Introduction (45)

Cyberinfrastructure for E-Applications

This introduction describes Cyberinfrastructure or e-infrastructure and its role in solving the electronic implementation of any problem where e-moreorlessanything is another term for moreorlessanything-Informatics and generalizes early discussion of e-Science and e-Business.

What is Cloud Computing: Introduction

Cloud Computing is introduced with an operational definition involving virtualization and efficient large data centers that can rent computers in an elastic fashion. The role of services is essential – it underlies capabilities being offered in the cloud. The four basic aaS’s – Software (SaaS), Platform (Paas), Infrastructure (IaaS) and Network (NaaS) – are introduced with Research aaS and other capabilities (for example Sensors aaS are discussed later) being built on top of these.

What and Why is Cloud Computing: Other Views I

This lesson contains 5 slides with diverse comments on ‘‘what is cloud computing’’ from the web.

Gartner’s Emerging Technology Landscape for Clouds and Big Data

This lesson gives Gartner’s projections around futures of cloud and Big data. We start with a review of hype charts and then go into detailed Gartner analyses of the Cloud and Big data areas. Big data itself is at the top of the hype and by definition predictions of doom are emerging. Before too much excitement sets in, note that spinach is above clouds and Big data in Google trends.

Simple Examples of use of Cloud Computing

This short lesson gives two examples of rather straightforward commercial applications of cloud computing. One is server consolidation for multiple Microsoft database applications and the second is the benefits of scale comparing gmail to multiple smaller installations. It ends with some fiscal comments.

Value of Cloud Computing

Some comments on fiscal value of cloud computing.

Resources

Software and Systems

We cover different views as to nature of architecture and application for Cloud Computing. Then we discuss cloud software for the cloud starting at virtual machine management (IaaS) and the broad Platform (middleware) capabilities with examples from Amazon and academic studies. We summarize the 21 layers and almost 300 software packages in the HPC-ABDS Software Stack explaining how they are used.

Presentation Software and Systems (32)

What is Cloud Computing

This lesson gives some general remark of cloud systems from an architecture and application perspective.

Introduction to Cloud Software Architecture: IaaS and PaaS I

We discuss cloud software for the cloud starting at virtual machine management (IaaS) and the broad Platform (middleware) capabilities with examples from Amazon and academic studies. We cover different views as to nature of architecture and application for Cloud Computing. Then we discuss cloud software for the cloud starting at virtual machine management (IaaS) and the broad Platform (middleware) capabilities with examples from Amazon and academic studies. We summarize the 21 layers and almost 300 software packages in the HPC-ABDS Software Stack explaining how they are used.

We discuss cloud software for the cloud starting at virtual machine management (IaaS) and the broad Platform (middleware) capabilities with examples from Amazon and academic studies. We cover different views as to nature of architecture and application for Cloud Computing. Then we discuss cloud software for the cloud starting at virtual machine management (IaaS) and the broad Platform (middleware) capabilities with examples from Amazon and academic studies. We summarize the 21 layers and almost 300 software packages in the HPC-ABDS Software Stack explaining how they are used.

Using the HPC-ABDS Software Stack

Using the HPC-ABDS Software Stack.

Resources

Architectures, Applications and Systems

We start with a discussion of Cloud (Data Center) Architectures with physical setup, Green Computing issues and software models. We summarize a 2014 Gartner analysis of Cloud computing providers. This is followed by applications on the cloud including data intensive problems, comparison with high performance computing, science clouds and the Internet of Things. Remarks on Security, Fault Tolerance and Synchronicity issues in cloud follow.

scroll: Architectures (64)

Cloud (Data Center) Architectures

Some remarks on what it takes to build (in software) a cloud ecosystem, and why clouds are the data center of the future are followed by pictures and discussions of several data centers from Microsoft (mainly) and Google. The role of containers is stressed as part of modular data centers that trade scalability for fault tolerance. Sizes of cloud centers and supercomputers are discussed as is “green” computing.

Analysis of Major Cloud Providers

Gartner 2014 Analysis of leading cloud providers.

Use of Dropbox, iCloud, Box etc.

Cloud Applications I

This short lesson discusses the need for security and issues in its implementation. Clouds trade scalability for greater possibility of faults but here clouds offer good support for recovery from faults. We discuss both storage and program fault tolerance noting that parallel computing is especially sensitive to faults as a fault in one task will impact all other tasks in the parallel job.

Science Clouds

Science Applications and Internet of Things.

Security

This short lesson discusses the need for security and issues in its implementation.

Comments on Fault Tolerance and Synchronicity Constraints

Clouds trade scalability for greater possibility of faults but here clouds offer good support for recovery from faults. We discuss both storage and program fault tolerance noting that parallel computing is especially sensitive to faults as a fault in one task will impact all other tasks in the parallel job.

Resources

Data Systems

We describe the way users and data interact with a cloud system. The unit concludes with the treatment of data in the cloud from an architecture perspective and Big Data Processing from an application perspective with commercial examples including eBay.

Presentation Data Systems (49)

The 10 Interaction scenarios (access patterns) I

The next 3 lessons describe the way users and data interact with the system.

The 10 Interaction scenarios. Science Examples

This lesson describes the way users and data interact with the system for some science examples.

Remaining general access patterns

This lesson describe the way users and data interact with the system for the final set of examples.

Video Access Patterns (11:36)

Data in the Cloud

Databases, File systems, Object Stores and NOSQL are discussed and compared. The way to build a modern data repository in the cloud is introduced.

Video Data in the Cloud (10:24)

Applications Processing Big Data

This lesson collects remarks on Big data processing from several sources: Berkeley, Teradata, IBM, Oracle and eBay with architectures and application opportunities.

Video Processing Big Data (8:45)

Resources

3 - e-Commerce and LifeStyle

e-Commerce and LifeStyle

Recommender systems operate under the hood of such widely recognized sites as Amazon, eBay, Monster and Netflix where everything is a recommendation. This involves a symbiotic relationship between vendor and buyer whereby the buyer provides the vendor with information about their preferences, while the vendor then offers recommendations tailored to match their needs. Kaggle competitions h improve the success of the Netflix and other recommender systems. Attention is paid to models that are used to compare how changes to the systems affect their overall performance. It is interesting that the humble ranking has become such a dominant driver of the world’s economy. More examples of recommender systems are given from Google News, Retail stores and in depth Yahoo! covering the multi-faceted criteria used in deciding recommendations on web sites.

The formulation of recommendations in terms of points in a space or bag is given where bags of item properties, user properties, rankings and users are useful. Detail is given on basic principles behind recommender systems: user-based collaborative filtering, which uses similarities in user rankings to predict their interests, and the Pearson correlation, used to statistically quantify correlations between users viewed as points in a space of items. Items are viewed as points in a space of users in item-based collaborative filtering. The Cosine Similarity is introduced, the difference between implicit and explicit ratings and the k Nearest Neighbors algorithm. General features like the curse of dimensionality in high dimensions are discussed. A simple Python k Nearest Neighbor code and its application to an artificial data set in 3 dimensions is given. Results are visualized in Matplotlib in 2D and with Plotviz in 3D. The concept of a training and a testing set are introduced with training set pre labeled. Recommender system are used to discuss clustering with k-means based clustering methods used and their results examined in Plotviz. The original labelling is compared to clustering results and extension to 28 clusters given. General issues in clustering are discussed including local optima, the use of annealing to avoid this and value of heuristic algorithms.

Recommender Systems

We introduce Recommender systems as an optimization technology used in a variety of applications and contexts online. They operate in the background of such widely recognized sites as Amazon, eBay, Monster and Netflix where everything is a recommendation. This involves a symbiotic relationship between vendor and buyer whereby the buyer provides the vendor with information about their preferences, while the vendor then offers recommendations tailored to match their needs, to the benefit of both.

There follows an exploration of the Kaggle competition site, other recommender systems and Netflix, as well as competitions held to improve the success of the Netflix recommender system. Finally attention is paid to models that are used to compare how changes to the systems affect their overall performance. It is interesting how the humble ranking has become such a dominant driver of the world’s economy.

Presentation Lifestyle Recommender (45)

Recommender Systems as an Optimization Problem

We define a set of general recommender systems as matching of items to people or perhaps collections of items to collections of people where items can be other people, products in a store, movies, jobs, events, web pages etc. We present this as “yet another optimization problem”.

Video Recommender Systems I (8:06)

Recommender Systems Introduction

We give a general discussion of recommender systems and point out that they are particularly valuable in long tail of tems (to be recommended) that are not commonly known. We pose them as a rating system and relate them to information retrieval rating systems. We can contrast recommender systems based on user profile and context; the most familiar collaborative filtering of others ranking; item properties; knowledge and hybrid cases mixing some or all of these.

Video Recommender Systems Introduction (12:56)

Kaggle Competitions

We look at Kaggle competitions with examples from web site. In particular we discuss an Irvine class project involving ranking jokes.

Video Kaggle Competitions: (3:36)

Warning Please not that we typically do not accept any projects using kaggle data for this classes. This class is not about winning a kaggle competition and if done wrong it does not fullfill the minimum requiremnt for this class. Please consult with the instructor.

Examples of Recommender Systems

We go through a list of 9 recommender systems from the same Irvine class.

Video Examples of Recommender Systems (1:00)

Netflix on Recommender Systems

We summarize some interesting points from a tutorial from Netflix for whom everything is a recommendation. Rankings are given in multiple categories and categories that reflect user interests are especially important. Criteria used include explicit user preferences, implicit based on ratings and hybrid methods as well as freshness and diversity. Netflix tries to explain the rationale of its recommendations. We give some data on Netflix operations and some methods used in its recommender systems. We describe the famous Netflix Kaggle competition to improve its rating system. The analogy to maximizing click through rate is given and the objectives of optimization are given.

Video Netflix on Recommender Systems (14:20)

Next we go through Netflix’s methodology in letting data speak for itself in optimizing the recommender engine. An example iis given on choosing self produced movies. A/B testing is discussed with examples showing how testing does allow optimizing of sophisticated criteria. This lesson is concluded by comments on Netflix technology and the full spectrum of issues that are involved including user interface, data, AB testing, systems and architectures. We comment on optimizing for a household rather than optimizing for individuals in household.

Video Consumer Data Science (13:04)

Other Examples of Recommender Systems

We continue the discussion of recommender systems and their use in e-commerce. More examples are given from Google News, Retail stores and in depth Yahoo! covering the multi-faceted criteria used in deciding recommendations on web sites. Then the formulation of recommendations in terms of points in a space or bag is given.

Here bags of item properties, user properties, rankings and users are useful. Then we go into detail on basic principles behind recommender systems: user-based collaborative filtering, which uses similarities in user rankings to predict their interests, and the Pearson correlation, used to statistically quantify correlations between users viewed as points in a space of items.

Presentation Lifestyle Recommender (49)

We start with a quick recap of recommender systems from previous unit; what they are with brief examples.

Video Recap and Examples of Recommender Systems (5:48)

Examples of Recommender Systems

We give 2 examples in more detail: namely Google News and Markdown in Retail.

Video Examples of Recommender Systems (8:34)

Recommender Systems in Yahoo Use Case Example

We describe in greatest detail the methods used to optimize Yahoo web sites. There are two lessons discussing general approach and a third lesson examines a particular personalized Yahoo page with its different components. We point out the different criteria that must be blended in making decisions; these criteria include analysis of what user does after a particular page is clicked; is the user satisfied and cannot that we quantified by purchase decisions etc. We need to choose Articles, ads, modules, movies, users, updates, etc to optimize metrics such as relevance score, CTR, revenue, engagement.These lesson stress that if though we have big data, the recommender data is sparse. We discuss the approach that involves both batch (offline) and on-line (real time) components.

Video Recap of Recommender Systems II (8:46)

Video Recap of Recommender Systems III (10:48)

Video Case Study of Recommender systems (3:21)

User-based nearest-neighbor collaborative filtering

Collaborative filtering is a core approach to recommender systems. There is user-based and item-based collaborative filtering and here we discuss the user-based case. Here similarities in user rankings allow one to predict their interests, and typically this quantified by the Pearson correlation, used to statistically quantify correlations between users.

Video User-based nearest-neighbor collaborative filtering I (7:20)

Video User-based nearest-neighbor collaborative filtering II (7:29)

Vector Space Formulation of Recommender Systems

We go through recommender systems thinking of them as formulated in a funny vector space. This suggests using clustering to make recommendations.

Video Vector Space Formulation of Recommender Systems new (9:06)

Resources

Item-based Collaborative Filtering and its Technologies

We move on to item-based collaborative filtering where items are viewed as points in a space of users. The Cosine Similarity is introduced, the difference between implicit and explicit ratings and the k Nearest Neighbors algorithm. General features like the curse of dimensionality in high dimensions are discussed.

Presentation Lifestyle Filtering (18)

Item-based Collaborative Filtering

We covered user-based collaborative filtering in the previous unit. Here we start by discussing memory-based real time and model based offline (batch) approaches. Now we look at item-based collaborative filtering where items are viewed in the space of users and the cosine measure is used to quantify distances. WE discuss optimizations and how batch processing can help. We discuss different Likert ranking scales and issues with new items that do not have a significant number of rankings.

Video Item Based Filtering (11:18)

Video k Nearest Neighbors and High Dimensional Spaces (7:16)

k-Nearest Neighbors and High Dimensional Spaces

We define the k Nearest Neighbor algorithms and present the Python software but do not use it. We give examples from Wikipedia and describe performance issues. This algorithm illustrates the curse of dimensionality. If items were a real vectors in a low dimension space, there would be faster solution methods.

Video k Nearest Neighbors and High Dimensional Spaces (10:03)

Recommender Systems - K-Neighbors

Next we provide some sample Python code for the k Nearest Neighbor and its application to an artificial data set in 3 dimensions. Results are visualized in Matplotlib in 2D and with Plotviz in 3D. The concept of training and testing sets are introduced with training set pre-labelled. This lesson is adapted from the Python k Nearest Neighbor code found on the web associated with a book by Harrington on Machine Learning [??]. There are two data sets. First we consider a set of 4 2D vectors divided into two categories (clusters) and use k=3 Nearest Neighbor algorithm to classify 3 test points. Second we consider a 3D dataset that has already been classified and show how to normalize. In this lesson we just use Matplotlib to give 2D plots.

The lesson goes through an example of using k NN classification algorithm by dividing dataset into 2 subsets. One is training set with initial classification; the other is test point to be classified by k=3 NN using training set. The code records fraction of points with a different classification from that input. One can experiment with different sizes of the two subsets. The Python implementation of algorithm is analyzed in detail.

Plotviz

The clustering methods are used and their results examined in Plotviz. The original labelling is compared to clustering results and extension to 28 clusters given. General issues in clustering are discussed including local optima, the use of annealing to avoid this and value of heuristic algorithms.

Files

Resources k-means

4 - Health Informatics

Health Informatics

Presentation Health Informatics (131)

This section starts by discussing general aspects of Big Data and Health including data sizes, different areas including genomics, EBI, radiology and the Quantified Self movement. We review current state of health care and trends associated with it including increased use of Telemedicine. We summarize an industry survey by GE and Accenture and an impressive exemplar Cloud-based medicine system from Potsdam. We give some details of big data in medicine. Some remarks on Cloud computing and Health focus on security and privacy issues.

We survey an April 2013 McKinsey report on the Big Data revolution in US health care; a Microsoft report in this area and a European Union report on how Big Data will allow patient centered care in the future. Examples are given of the Internet of Things, which will have great impact on health including wearables. A study looks at 4 scenarios for healthcare in 2032. Two are positive, one middle of the road and one negative. The final topic is Genomics, Proteomics and Information Visualization.

Big Data and Health

This lesson starts with general aspects of Big Data and Health including listing subareas where Big data important. Data sizes are given in radiology, genomics, personalized medicine, and the Quantified Self movement, with sizes and access to European Bioinformatics Institute.

Video Big Data and Health (10:02)

Status of Healthcare Today

This covers trends of costs and type of healthcare with low cost genomes and an aging population. Social media and government Brain initiative.

Video Status of Healthcare Today (16:09)

Telemedicine (Virtual Health)

This describes increasing use of telemedicine and how we tried and failed to do this in 1994.

Video Telemedicine (8:21)

Medical Big Data in the Clouds

An impressive exemplar Cloud-based medicine system from Potsdam.

Video Medical Big Data in the Clouds (15:02)

Medical image Big Data

Video Medical Image Big Data (6:33)

Clouds and Health

Video Clouds and Health (4:35)

McKinsey Report on the big-data revolution in US health care

This lesson covers 9 aspects of the McKinsey report. These are the convergence of multiple positive changes has created a tipping point for

innovation; Primary data pools are at the heart of the big data revolution in healthcare; Big data is changing the paradigm: these are the value pathways; Applying early successes at scale could reduce US healthcare costs by $300 billion to $450 billion; Most new big-data applications target consumers and providers across pathways; Innovations are weighted towards influencing individual decision-making levers; Big data innovations use a range of public, acquired, and proprietary data

types; Organizations implementing a big data transformation should provide the leadership required for the associated cultural transformation; Companies must develop a range of big data capabilities.

Video McKinsey Report (14:53)

Microsoft Report on Big Data in Health

This lesson identifies data sources as Clinical Data, Pharma & Life Science Data, Patient & Consumer Data, Claims & Cost Data and Correlational Data. Three approaches are Live data feed, Advanced analytics and Social analytics.

Video Microsoft Report on Big Data in Health (2:26)

EU Report on Redesigning health in Europe for 2020

This lesson summarizes an EU Report on Redesigning health in Europe for 2020. The power of data is seen as a lever for change in My Data, My decisions; Liberate the data; Connect up everything; Revolutionize health; and Include Everyone removing the current correlation between health and wealth.

Video EU Report on Redesigning health in Europe for 2020 (5:00)

Medicine and the Internet of Things

The Internet of Things will have great impact on health including telemedicine and wearables. Examples are given.

Video Medicine and the Internet of Things (8:17)

Extrapolating to 2032

A study looks at 4 scenarios for healthcare in 2032. Two are positive, one middle of the road and one negative.

Video Extrapolating to 2032 (15:13)

Genomics, Proteomics and Information Visualization

A study of an Azure application with an Excel frontend and a cloud BLAST backend starts this lesson. This is followed by a big data analysis of personal genomics and an analysis of a typical DNA sequencing analytics pipeline. The Protein Sequence Universe is defined and used to motivate Multi dimensional Scaling MDS. Sammon’s method is defined and its use illustrated by a metagenomics example. Subtleties in use of MDS include a monotonic mapping of the dissimilarity function. The application to the COG Proteomics dataset is discussed. We note that the MDS approach is related to the well known chisq method and some aspects of nonlinear minimization of chisq (Least Squares) are discussed.

Video Genomics, Proteomics and Information Visualization (6:56)

Next we continue the discussion of the COG Protein Universe introduced in the last lesson. It is shown how Proteomics clusters are clearly seen in the Universe browser. This motivates a side remark on different clustering methods applied to metagenomics. Then we discuss the Generative Topographic Map GTM method that can be used in dimension reduction when original data is in a metric space and is in this case faster than MDS as GTM computational complexity scales like N not N squared as seen in MDS.

Examples are given of GTM including an application to topic models in Information Retrieval. Indiana University has developed a deterministic annealing improvement of GTM. 3 separate clusterings are projected for visualization and show very different structure emphasizing the importance of visualizing results of data analytics. The final slide shows an application of MDS to generate and visualize phylogenetic trees.

\TODO{These two videos need to be uploaded to youtube} Video Genomics, Proteomics and Information Visualization I (10:33)

Video Genomics, Proteomics and Information Visualization: II (7:41)

Presentation Proteomics and Information Visualization (131)

Resources

5 - Overview of Data Science

Overview of Data Science

What is Big Data, Data Analytics and X-Informatics?

We start with X-Informatics and its rallying cry. The growing number of jobs in data science is highlighted. The first unit offers a look at the phenomenon described as the Data Deluge starting with its broad features. Data science and the famous DIKW (Data to Information to Knowledge to Wisdom) pipeline are covered. Then more detail is given on the flood of data from Internet and Industry applications with eBay and General Electric discussed in most detail.

In the next unit, we continue the discussion of the data deluge with a focus on scientific research. He takes a first peek at data from the Large Hadron Collider considered later as physics Informatics and gives some biology examples. He discusses the implication of data for the scientific method which is changing with the data-intensive methodology joining observation, theory and simulation as basic methods. Two broad classes of data are the long tail of sciences: many users with individually modest data adding up to a lot; and a myriad of Internet connected devices – the Internet of Things.

We give an initial technical overview of cloud computing as pioneered by companies like Amazon, Google and Microsoft with new centers holding up to a million servers. The benefits of Clouds in terms of power consumption and the environment are also touched upon, followed by a list of the most critical features of Cloud computing with a comparison to supercomputing. Features of the data deluge are discussed with a salutary example where more data did better than more thought. Then comes Data science and one part of it ~~ data analytics ~~ the large algorithms that crunch the big data to give big wisdom. There are many ways to describe data science and several are discussed to give a good composite picture of this emerging field.

Data Science generics and Commercial Data Deluge

We start with X-Informatics and its rallying cry. The growing number of jobs in data science is highlighted. This unit offers a look at the phenomenon described as the Data Deluge starting with its broad features. Then he discusses data science and the famous DIKW (Data to Information to Knowledge to Wisdom) pipeline. Then more detail is given on the flood of data from Internet and Industry applications with eBay and General Electric discussed in most detail.

Presentation Commercial Data Deluge (45)

What is X-Informatics and its Motto

This discusses trends that are driven by and accompany Big data. We give some key terms including data, information, knowledge, wisdom, data analytics and data science. We discuss how clouds running Data Analytics Collaboratively processing Big Data can solve problems in X-Informatics. We list many values of X you can defined in various activities across the world.

Jobs

Big data is especially important as there are some many related jobs. We illustrate this for both cloud computing and data science from reports by Microsoft and the McKinsey institute respectively. We show a plot from LinkedIn showing rapid increase in the number of data science and analytics jobs as a function of time.

Data Deluge: General Structure

We look at some broad features of the data deluge starting with the size of data in various areas especially in science research. We give examples from real world of the importance of big data and illustrate how it is integrated into an enterprise IT architecture. We give some views as to what characterizes Big data and why data science is a science that is needed to interpret all the data.

Data Science: Process

We stress the DIKW pipeline: Data becomes information that becomes knowledge and then wisdom, policy and decisions. This pipeline is illustrated with Google maps and we show how complex the ecosystem of data, transformations (filters) and its derived forms is.

Data Deluge: Internet

We give examples of Big data from the Internet with Tweets, uploaded photos and an illustration of the vitality and size of many commodity applications.

Data Deluge: Business

We give examples including the Big data that enables wind farms, city transportation, telephone operations, machines with health monitors, the banking, manufacturing and retail industries both online and offline in shopping malls. We give examples from ebay showing how analytics allowing them to refine and improve the customer experiences.

Resources

Data Deluge and Scientific Applications and Methodology

Overview of Data Science

We continue the discussion of the data deluge with a focus on scientific research. He takes a first peek at data from the Large Hadron Collider considered later as physics Informatics and gives some biology examples. He discusses the implication of data for the scientific method which is changing with the data-intensive methodology joining observation, theory and simulation as basic methods. We discuss the long tail of sciences; many users with individually modest data adding up to a lot. The last lesson emphasizes how everyday devices ~~ the Internet of Things ~~ are being used to create a wealth of data.

Presentation Methodology (22)

Science and Research

We look into more big data examples with a focus on science and research. We give astronomy, genomics, radiology, particle physics and discovery of Higgs particle (Covered in more detail in later lessons), European Bioinformatics Institute and contrast to Facebook and Walmart.

Implications for Scientific Method

We discuss the emergencies of a new fourth methodology for scientific research based on data driven inquiry. We contrast this with third ~~ computation or simulation based discovery - methodology which emerged itself some 25 years ago.

Long Tail of Science

There is big science such as particle physics where a single experiment has 3000 people collaborate!.Then there are individual investigators who do not generate a lot of data each but together they add up to Big data.

Internet of Things

A final category of Big data comes from the Internet of Things where lots of small devices ~~ smart phones, web cams, video games collect and disseminate data and are controlled and coordinated in the cloud.

Resources

Clouds and Big Data Processing; Data Science Process and Analytics

Overview of Data Science

We give an initial technical overview of cloud computing as pioneered by companies like Amazon, Google and Microsoft with new centers holding up to a million servers. The benefits of Clouds in terms of power consumption and the environment are also touched upon, followed by a list of the most critical features of Cloud computing with a comparison to supercomputing.

He discusses features of the data deluge with a salutary example where more data did better than more thought. He introduces data science and one part of it ~~ data analytics ~~ the large algorithms that crunch the big data to give big wisdom. There are many ways to describe data science and several are discussed to give a good composite picture of this emerging field.

Presentation Clouds (35)

Clouds

We describe cloud data centers with their staggering size with up to a million servers in a single data center and centers built modularly from shipping containers full of racks. The benefits of Clouds in terms of power consumption and the environment are also touched upon, followed by a list of the most critical features of Cloud computing and a comparison to supercomputing.

Aspect of Data Deluge

Data, Information, intelligence algorithms, infrastructure, data structure, semantics and knowledge are related. The semantic web and Big data are compared. We give an example where “More data usually beats better algorithms”. We discuss examples of intelligent big data and list 8 different types of data deluge

Data Science Process

We describe and critique one view of the work of a data scientists. Then we discuss and contrast 7 views of the process needed to speed data through the DIKW pipeline.

Data Analytics

Presentation Data Analytics (30) We stress the importance of data analytics givi ng examples from several fields. We note that better analytics is as important as better computing and storage capability. In the second video we look at High Performance Computing in Science and Engineering: the Tree and the Fruit.

Resources

6 - Physics

Physics

This section starts by describing the LHC accelerator at CERN and evidence found by the experiments suggesting existence of a Higgs Boson. The huge number of authors on a paper, remarks on histograms and Feynman diagrams is followed by an accelerator picture gallery. The next unit is devoted to Python experiments looking at histograms of Higgs Boson production with various forms of shape of signal and various background and with various event totals. Then random variables and some simple principles of statistics are introduced with explanation as to why they are relevant to Physics counting experiments. The unit introduces Gaussian (normal) distributions and explains why they seen so often in natural phenomena. Several Python illustrations are given. Random Numbers with their Generators and Seeds lead to a discussion of Binomial and Poisson Distribution. Monte-Carlo and accept-reject methods. The Central Limit Theorem concludes discussion.

Looking for Higgs Particles

Bumps in Histograms, Experiments and Accelerators

This unit is devoted to Python and Java experiments looking at histograms of Higgs Boson production with various forms of shape of signal and various background and with various event totals. The lectures use Python but use of Java is described.

  • Presentation Higgs (20)

  • <{gitcode}/physics/mr-higgs/higgs-classI-sloping.py>

Particle Counting

We return to particle case with slides used in introduction and stress that particles often manifested as bumps in histograms and those bumps need to be large enough to stand out from background in a statistically significant fashion.

We give a few details on one LHC experiment ATLAS. Experimental physics papers have a staggering number of authors and quite big budgets. Feynman diagrams describe processes in a fundamental fashion.

Experimental Facilities

We give a few details on one LHC experiment ATLAS. Experimental physics papers have a staggering number of authors and quite big budgets. Feynman diagrams describe processes in a fundamental fashion.

This lesson gives a small picture gallery of accelerators. Accelerators, detection chambers and magnets in tunnels and a large underground laboratory used fpr experiments where you need to be shielded from background like cosmic rays.

Resources

Looking for Higgs Particles: Python Event Counting for Signal and Background (Part 2)

This unit is devoted to Python experiments looking at histograms of Higgs Boson production with various forms of shape of signal and various background and with various event totals.

Files:

  • <{gitcode}/physics/mr-higgs/higgs-classI-sloping.py>
  • <{gitcode}/physics/number-theory/higgs-classIII.py>
  • <{gitcode}/physics/mr-higgs/higgs-classII-uniform.py>

Event Counting

We define event counting data collection environments. We discuss the python and Java code to generate events according to a particular scenario (the important idea of Monte Carlo data). Here a sloping background plus either a Higgs particle generated similarly to LHC observation or one observed with better resolution (smaller measurement error).

Monte Carlo

This uses Monte Carlo data both to generate data like the experimental observations and explore effect of changing amount of data and changing measurement resolution for Higgs.

Resources

Random Variables, Physics and Normal Distributions

We introduce random variables and some simple principles of statistics and explains why they are relevant to Physics counting experiments. The unit introduces Gaussian (normal) distributions and explains why they seen so often in natural phenomena. Several Python illustrations are given. Java is currently not available in this unit.

  • Presentation Higgs (39)
  • <{gitcode}/physics/number-theory/higgs-classIII.py>

Statistics Overview and Fundamental Idea: Random Variables

We go through the many different areas of statistics covered in the Physics unit. We define the statistics concept of a random variable.

Physics and Random Variables

We describe the DIKW pipeline for the analysis of this type of physics experiment and go through details of analysis pipeline for the LHC ATLAS experiment. We give examples of event displays showing the final state particles seen in a few events. We illustrate how physicists decide whats going on with a plot of expected Higgs production experimental cross sections (probabilities) for signal and background.

Statistics of Events with Normal Distributions

We introduce Poisson and Binomial distributions and define independent identically distributed (IID) random variables. We give the law of large numbers defining the errors in counting and leading to Gaussian distributions for many things. We demonstrate this in Python experiments.

Gaussian Distributions

We introduce the Gaussian distribution and give Python examples of the fluctuations in counting Gaussian distributions.

Using Statistics

We discuss the significance of a standard deviation and role of biases and insufficient statistics with a Python example in getting incorrect answers.

Resources

Random Numbers, Distributions and Central Limit Theorem

We discuss Random Numbers with their Generators and Seeds. It introduces Binomial and Poisson Distribution. Monte-Carlo and accept-reject methods are discussed. The Central Limit Theorem and Bayes law concludes discussion. Python and Java (for student - not reviewed in class) examples and Physics applications are given.

Files:

  • <{gitcode}/physics/calculated-dice-roll/higgs-classIV-seeds.py>

Generators and Seeds

We define random numbers and describe how to generate them on the computer giving Python examples. We define the seed used to define to specify how to start generation.

Binomial Distribution

We define binomial distribution and give LHC data as an example of where this distribution valid.

Accept-Reject

We introduce an advanced method accept/reject for generating random variables with arbitrary distributions.

Monte Carlo Method

We define Monte Carlo method which usually uses accept/reject method in typical case for distribution.

Poisson Distribution

We extend the Binomial to the Poisson distribution and give a set of amusing examples from Wikipedia.

Central Limit Theorem

We introduce Central Limit Theorem and give examples from Wikipedia.

Interpretation of Probability: Bayes v. Frequency

This lesson describes difference between Bayes and frequency views of probability. Bayes’s law of conditional probability is derived and applied to Higgs example to enable information about Higgs from multiple channels and multiple experiments to be accumulated.

Resources

\TODO{integrate physics-references.bib}

SKA – Square Kilometer Array

Professor Diamond, accompanied by Dr. Rosie Bolton from the SKA Regional Centre Project gave a presentation at SC17 “into the deepest reaches of the observable universe as they describe the SKA’s international partnership that will map and study the entire sky in greater detail than ever before.”

A summary article about this effort is available at:

7 - Plotviz

Plotviz

NOTE: This an legacy application this has now been replaced by WebPlotViz which is a web browser based visualization tool which provides added functionality’s.

We introduce Plotviz, a data visualization tool developed at Indiana University to display 2 and 3 dimensional data. The motivation is that the human eye is very good at pattern recognition and can see structure in data. Although most Big data is higher dimensional than 3, all can be transformed by dimension reduction techniques to 3D. He gives several examples to show how the software can be used and what kind of data can be visualized. This includes individual plots and the manipulation of multiple synchronized plots.Finally, he describes the download and software dependency of Plotviz.

Using Plotviz Software for Displaying Point Distributions in 3D

We introduce Plotviz, a data visualization tool developed at Indiana University to display 2 and 3 dimensional data. The motivation is that the human eye is very good at pattern recognition and can see structure in data. Although most Big data is higher dimensional than 3, all can be transformed by dimension reduction techniques to 3D. He gives several examples to show how the software can be used and what kind of data can be visualized. This includes individual plots and the manipulation of multiple synchronized plots. Finally, he describes the download and software dependency of Plotviz.

Presentation Plotviz (34)

Files:

Motivation and Introduction to use

The motivation of Plotviz is that the human eye is very good at pattern recognition and can see structure in data. Although most Big data is higher dimensional than 3, all data can be transformed by dimension reduction techniques to 3D and one can check analysis like clustering and/or see structure missed in a computer analysis. The motivations shows some Cheminformatics examples. The use of Plotviz is started in slide 4 with a discussion of input file which is either a simple text or more features (like colors) can be specified in a rich XML syntax. Plotviz deals with points and their classification (clustering). Next the protein sequence browser in 3D shows the basic structure of Plotviz interface. The next two slides explain the core 3D and 2D manipulations respectively. Note all files used in examples are available to students.

Presentation Motivation (7:58)

Example of Use I: Cube and Structured Dataset

Initially we start with a simple plot of 8 points – the corners of a cube in 3 dimensions – showing basic operations such as size/color/labels and Legend of points. The second example shows a dataset (coming from GTM dimension reduction) with significant structure. This has .pviz and a .txt versions that are compared.

Presentation Example I (9:45)

Example of Use II: Proteomics and Synchronized Rotation

This starts with an examination of a sample of Protein Universe Browser showing how one uses Plotviz to look at different features of this set of Protein sequences projected to 3D. Then we show how to compare two datasets with synchronized rotation of a dataset clustered in 2 different ways; this dataset comes from k Nearest Neighbor discussion.

Presentation Proteomics and Synchronized Rotation (9:14)

Example of Use III: More Features and larger Proteomics Sample

This starts by describing use of Labels and Glyphs and the Default mode in Plotviz. Then we illustrate sophisticated use of these ideas to view a large Proteomics dataset.

Presentation Larger Proteomics Sample (8:37)

Example of Use IV: Tools and Examples

This lesson starts by describing the Plotviz tools and then sets up two examples – Oil Flow and Trading – described in PowerPoint. It finishes with the Plotviz viewing of Oil Flow data.

Presentation Plotviz I (10:17)

Example of Use V: Final Examples

This starts with Plotviz looking at Trading example introduced in previous lesson and then examines solvent data. It finishes with two large biology examples with 446K and 100K points and each with over 100 clusters. We finish remarks on Plotviz software structure and how to download. We also remind you that a picture is worth a 1000 words.

Video Plotviz II (14:58)

Resources

Download

8 - Practical K-Means, Map Reduce, and Page Rank for Big Data Applications and Analytics

Practical K-Means, Map Reduce, and Page Rank for Big Data Applications and Analytics

We use the K-means Python code in SciPy package to show real code for clustering. After a simple example we generate 4 clusters of distinct centers and various choice for sizes using Matplotlib tor visualization. We show results can sometimes be incorrect and sometimes make different choices among comparable solutions. We discuss the hill between different solutions and rationale for running K-means many times and choosing best answer. Then we introduce MapReduce with the basic architecture and a homely example. The discussion of advanced topics includes an extension to Iterative MapReduce from Indiana University called Twister and a generalized Map Collective model. Some measurements of parallel performance are given. The SciPy K-means code is modified to support a MapReduce execution style. This illustrates the key ideas of mappers and reducers. With appropriate runtime this code would run in parallel but here the parallel maps run sequentially. This simple 2 map version can be generalized to scalable parallelism. Python is used to Calculate PageRank from Web Linkage Matrix showing several different formulations of the basic matrix equations to finding leading eigenvector. The unit is concluded by a calculation of PageRank for general web pages by extracting the secret from Google.

Video K-Means I (11:42)

Video K-Means II (11:54)

K-means in Practice

We introduce the k means algorithm in a gentle fashion and describes its key features including dangers of local minima. A simple example from Wikipedia is examined.

We use the K-means Python code in SciPy package to show real code for clustering. After a simple example we generate 4 clusters of distinct centers and various choice for sizes using Matplotlib tor visualization. We show results can sometimes be incorrect and sometimes make different choices among comparable solutions. We discuss the hill between different solutions and rationale for running K-means many times and choosing best answer.

Files:

K-means in Python

We use the K-means Python code in SciPy package to show real code for clustering and applies it a set of 85 two dimensional vectors – officially sets of weights and heights to be clustered to find T-shirt sizes. We run through Python code with Matplotlib displays to divide into 2-5 clusters. Then we discuss Python to generate 4 clusters of varying sizes and centered at corners of a square in two dimensions. We formally give the K means algorithm better than before and make definition consistent with code in SciPy.

Analysis of 4 Artificial Clusters

We present clustering results on the artificial set of 1000 2D points described in previous lesson for 3 choices of cluster sizes small large and very large. We emphasize the SciPy always does 20 independent K means and takes the best result – an approach to avoiding local minima. We allow this number of independent runs to be changed and in particular set to 1 to generate more interesting erratic results. We define changes in our new K means code that also has two measures of quality allowed. The slides give many results of clustering into 2 4 6 and 8 clusters (there were only 4 real clusters). We show that the very small case has two very different solutions when clustered into two clusters and use this to discuss functions with multiple minima and a hill between them. The lesson has both discussion of already produced results in slides and interactive use of Python for new runs.

Parallel K-means

We modify the SciPy K-means code to support a MapReduce execution style and runs it in this short unit. This illustrates the key ideas of mappers and reducers. With appropriate runtime this code would run in parallel but here the parallel maps run sequentially. We stress that this simple 2 map version can be generalized to scalable parallelism.

Files:

PageRank in Practice

We use Python to Calculate PageRank from Web Linkage Matrix showing several different formulations of the basic matrix equations to finding leading eigenvector. The unit is concluded by a calculation of PageRank for general web pages by extracting the secret from Google.

Files:

Resources

9 - Radar

Radar

The changing global climate is suspected to have long-term effects on much of the world’s inhabitants. Among the various effects, the rising sea level will directly affect many people living in low-lying coastal regions. While the ocean-s thermal expansion has been the dominant contributor to rises in sea level, the potential contribution of discharges from the polar ice sheets in Greenland and Antarctica may provide a more significant threat due to the unpredictable response to the changing climate. The Radar-Informatics unit provides a glimpse in the processes fueling global climate change and explains what methods are used for ice data acquisitions and analysis.

Presentation Radar (58)

Introduction

This lesson motivates radar-informatics by building on previous discussions on why X-applications are growing in data size and why analytics are necessary for acquiring knowledge from large data. The lesson details three mosaics of a changing Greenland ice sheet and provides a concise overview to subsequent lessons by detailing explaining how other remote sensing technologies, such as the radar, can be used to sound the polar ice sheets and what we are doing with radar images to extract knowledge to be incorporated into numerical models.

Remote Sensing

This lesson explains the basics of remote sensing, the characteristics of remote sensors and remote sensing applications. Emphasis is on image acquisition and data collection in the electromagnetic spectrum.

Ice Sheet Science

This lesson provides a brief understanding on why melt water at the base of the ice sheet can be detrimental and why it’s important for sensors to sound the bedrock.

Global Climate Change

This lesson provides an understanding and the processes for the greenhouse effect, how warming effects the Polar Regions, and the implications of a rise in sea level.

Radio Overview

This lesson provides an elementary introduction to radar and its importance to remote sensing, especially to acquiring information about Greenland and Antarctica.

Radio Informatics

This lesson focuses on the use of sophisticated computer vision algorithms, such as active contours and a hidden markov model to support data analysis for extracting layers, so ice sheet models can accurately forecast future changes in climate.

10 - Sensors

Sensors

We start with the Internet of Things IoT giving examples like monitors of machine operation, QR codes, surveillance cameras, scientific sensors, drones and self driving cars and more generally transportation systems. We give examples of robots and drones. We introduce the Industrial Internet of Things IIoT and summarize surveys and expectations Industry wide. We give examples from General Electric. Sensor clouds control the many small distributed devices of IoT and IIoT. More detail is given for radar data gathered by sensors; ubiquitous or smart cities and homes including U-Korea; and finally the smart electric grid.

Presentation Sensor I(31)

Presentation Sensor II(44)

Internet of Things

There are predicted to be 24-50 Billion devices on the Internet by 2020; these are typically some sort of sensor defined as any source or sink of time series data. Sensors include smartphones, webcams, monitors of machine operation, barcodes, surveillance cameras, scientific sensors (especially in earth and environmental science), drones and self driving cars and more generally transportation systems. The lesson gives many examples of distributed sensors, which form a Grid that is controlled by a cloud.

Video Internet of Things (12:36)

Robotics and IoT

Examples of Robots and Drones.

Video Robotics and IoT Expectations (8:05)

Industrial Internet of Things

We summarize surveys and expectations Industry wide.

Video Industrial Internet of Things (24:02)

Sensor Clouds

We describe the architecture of a Sensor Cloud control environment and gives example of interface to an older version of it. The performance of system is measured in terms of processing latency as a function of number of involved sensors with each delivering data at 1.8 Mbps rate.

Video Sensor Clouds (4:40)

Earth/Environment/Polar Science data gathered by Sensors

This lesson gives examples of some sensors in the Earth/Environment/Polar Science field. It starts with material from the CReSIS polar remote sensing project and then looks at the NSF Ocean Observing Initiative and NASA’s MODIS or Moderate Resolution Imaging Spectroradiometer instrument on a satellite.

Video Earth/Environment/Polar Science data gathered by Sensors (4:58)

Ubiquitous/Smart Cities

For Ubiquitous/Smart cities we give two examples: Iniquitous Korea and smart electrical grids.

Video Ubiquitous/Smart Cities (1:44)

U-Korea (U=Ubiquitous)

Korea has an interesting positioning where it is first worldwide in broadband access per capita, e-government, scientific literacy and total working hours. However it is far down in measures like quality of life and GDP. U-Korea aims to improve the latter by Pervasive computing, everywhere, anytime i.e. by spreading sensors everywhere. The example of a ‘High-Tech Utopia’ New Songdo is given.

Video U-Korea (U=Ubiquitous) (2:49)

Smart Grid

The electrical Smart Grid aims to enhance USA’s aging electrical infrastructure by pervasive deployment of sensors and the integration of their measurement in a cloud or equivalent server infrastructure. A variety of new instruments include smart meters, power monitors, and measures of solar irradiance, wind speed, and temperature. One goal is autonomous local power units where good use is made of waste heat.

Video Smart Grid (6:04)

Resources

\TODO{These resources have not all been checked to see if they still exist this is currently in progress}

11 - Sports

Sports

Sports sees significant growth in analytics with pervasive statistics shifting to more sophisticated measures. We start with baseball as game is built around segments dominated by individuals where detailed (video/image) achievement measures including PITCHf/x and FIELDf/x are moving field into big data arena. There are interesting relationships between the economics of sports and big data analytics. We look at Wearables and consumer sports/recreation. The importance of spatial visualization is discussed. We look at other Sports: Soccer, Olympics, NFL Football, Basketball, Tennis and Horse Racing.

Basic Sabermetrics

This unit discusses baseball starting with the movie Moneyball and the 2002-2003 Oakland Athletics. Unlike sports like basketball and soccer, most baseball action is built around individuals often interacting in pairs. This is much easier to quantify than many player phenomena in other sports. We discuss Performance-Dollar relationship including new stadiums and media/advertising. We look at classic baseball averages and sophisticated measures like Wins Above Replacement.

Presentation Overview (40)

Introduction and Sabermetrics (Baseball Informatics) Lesson

Introduction to all Sports Informatics, Moneyball The 2002-2003 Oakland Athletics, Diamond Dollars economic model of baseball, Performance - Dollar relationship, Value of a Win.

Video Introduction and Sabermetrics (Baseball Informatics) Lesson (31:4)

Basic Sabermetrics

Different Types of Baseball Data, Sabermetrics, Overview of all data, Details of some statistics based on basic data, OPS, wOBA, ERA, ERC, FIP, UZR.

Video Basic Sabermetrics (26:53)

Wins Above Replacement

Wins above Replacement WAR, Discussion of Calculation, Examples, Comparisons of different methods, Coefficient of Determination, Another, Sabermetrics Example, Summary of Sabermetrics.

Video Wins Above Replacement (30:43)

Advanced Sabermetrics

This unit discusses ‘advanced sabermetrics’ covering advances possible from using video from PITCHf/X, FIELDf/X, HITf/X, COMMANDf/X and MLBAM.

Presentation Sporta II (41)

Pitching Clustering

A Big Data Pitcher Clustering method introduced by Vince Gennaro, Data from Blog and video at 2013 SABR conference.

Video Pitching Clustering (20:59)

Pitcher Quality

Results of optimizing match ups, Data from video at 2013 SABR conference.

Video Pitcher Quality (10:02)

PITCHf/X

Examples of use of PITCHf/X.

Video PITCHf/X (10:39)

Other Video Data Gathering in Baseball

FIELDf/X, MLBAM, HITf/X, COMMANDf/X.

Video Other Video Data Gathering in Baseball (18:5) Other Sports

We look at Wearables and consumer sports/recreation. The importance of spatial visualization is discussed. We look at other Sports: Soccer, Olympics, NFL Football, Basketball, Tennis and Horse Racing.

Presentation Sport Sports III (44)

Wearables

Consumer Sports, Stake Holders, and Multiple Factors.

Video Wearables (22:2)

Soccer and the Olympics

Soccer, Tracking Players and Balls, Olympics.

Video Soccer and the Olympics (8:28)

Spatial Visualization in NFL and NBA

NFL, NBA, and Spatial Visualization.

Video Spatial Visualization in NFL and NBA (15:19)

Tennis and Horse Racing

Tennis, Horse Racing, and Continued Emphasis on Spatial Visualization.

Video Tennis and Horse Racing (8:52)

Resources

\TODO{These resources have not all been checked to see if they still exist this is currently in progress}

12 - Statistics

Statistics

We assume that you are familiar with elementary statistics including

  • mean, minimum, maximum
  • standard deviation
  • probability
  • distribution
  • frequency distribution
  • Gaussian distribution
  • bell curve
  • standard normal probabilities
  • tables (z table)
  • Regression
  • Correlation

Some of these terms are explained in various sections throughout our application discussion. This includes especially the Physics section. However these terms are so elementary that any undergraduate or highschool book will provide you with a good introduction.

It is expected from you to identify these terms and you can contribute to this section with non plagiarized subsections explaining these topics for credit.

No Topics identified by a :?: can be contributed by students. If you are interested, use piazza for announcing your willingness to do so.

Mean, minimum, maximum:

No Question

Standard deviation:

No Question

Probability:

No Question

Distribution:

No Question

Frequency distribution:

No Question

Gaussian distribution:

No Question

Bell curve:

No Question

Standard normal probabilities:

No Question

Tables (z-table):

No Question

Regression:

No Question

Correlation:

No Question

Exercise

E.Statistics.1:

Pick a term from the previous list and define it while not plagiarizing. Create a pull request. Coordinate on piazza as to not duplicate someone else’s contribution. Also look into outstanding pull requests.

E.Statistics.2:

Pick a term from the previous list and develop a python program demonstrating it and create a pull request for a contribution into the examples directory. Make links to the github location. Coordinate on piazza as to not duplicate someone else’s contribution. Also look into outstanding pull requests.

13 - Web Search and Text Mining

Web Search and Text Mining

This section starts with an overview of data mining and puts our study of classification, clustering and exploration methods in context. We examine the problem to be solved in web and text search and note the relevance of history with libraries, catalogs and concordances. An overview of web search is given describing the continued evolution of search engines and the relation to the field of Information.

The importance of recall, precision and diversity is discussed. The important Bag of Words model is introduced and both Boolean queries and the more general fuzzy indices. The important vector space model and revisiting the Cosine Similarity as a distance in this bag follows. The basic TF-IDF approach is dis cussed. Relevance is discussed with a probabilistic model while the distinction between Bayesian and frequency views of probability distribution completes this unit.

We start with an overview of the different steps (data analytics) in web search and then goes key steps in detail starting with document preparation. An inverted index is described and then how it is prepared for web search. The Boolean and Vector Space approach to query processing follow. This is followed by Link Structure Analysis including Hubs, Authorities and PageRank. The application of PageRank ideas as reputation outside web search is covered. The web graph structure, crawling it and issues in web advertising and search follow. The use of clustering and topic models completes the section.

Web Search and Text Mining

The unit starts with the web with its size, shape (coming from the mutual linkage of pages by URL’s) and universal power laws for number of pages with particular number of URL’s linking out or in to page. Information retrieval is introduced and compared to web search. A comparison is given between semantic searches as in databases and the full text search that is base of Web search. The origin of web search in libraries, catalogs and concordances is summarized. DIKW – Data Information Knowledge Wisdom – model for web search is discussed. Then features of documents, collections and the important Bag of Words representation. Queries are presented in context of an Information Retrieval architecture. The method of judging quality of results including recall, precision and diversity is described. A time line for evolution of search engines is given.

Boolean and Vector Space models for query including the cosine similarity are introduced. Web Crawlers are discussed and then the steps needed to analyze data from Web and produce a set of terms. Building and accessing an inverted index is followed by the importance of term specificity and how it is captured in TF-IDF. We note how frequencies are converted into belief and relevance.

Presentation Web Search and Text Mining (56)

The Problem

Video Text Mining (9:56)

This lesson starts with the web with its size, shape (coming from the mutual linkage of pages by URL’s) and universal power laws for number of pages with particular number of URL’s linking out or in to page.

Information Retrieval

Video Information Retrieval (6:06)

Information retrieval is introduced A comparison is given between semantic searches as in databases and the full text search that is base of Web search. The ACM classification illustrates potential complexity of ontologies. Some differences between web search and information retrieval are given.

History

Video Web Search History (5:48)

The origin of web search in libraries, catalogs and concordances is summarized.

Key Fundamental Principles

Video Principles (9:30)

This lesson describes the DIKW – Data Information Knowledge Wisdom – model for web search. Then it discusses documents, collections and the important Bag of Words representation.

Information Retrieval (Web Search) Components

Video Fundamental Principals of Web Search (5:06)

This describes queries in context of an Information Retrieval architecture. The method of judging quality of results including recall, precision and diversity is described.

Search Engines

Video Search Engines (3:08)

This short lesson describes a time line for evolution of search engines. The first web search approaches were directly built on Information retrieval but in 1998 the field was changed when Google was founded and showed the importance of URL structure as exemplified by PageRank.

Boolean and Vector Space Models

Video Boolean and Vector Space Model (6:17)

This lesson describes the Boolean and Vector Space models for query including the cosine similarity.

Web crawling and Document Preparation

Video Web crawling and Document Preparation (4:55)

This describes a Web Crawler and then the steps needed to analyze data from Web and produce a set of terms.

Indices

Video Indices (5:44)

This lesson describes both building and accessing an inverted index. It describes how phrases are treated and gives details of query structure from some early logs.

TF-IDF and Probabilistic Models

Video TF-IDF and Probabilistic Models (3:57)

It describes the importance of term specificity and how it is captured in TF-IDF. It notes how frequencies are converted into belief and relevance.

Topics in Web Search and Text Mining

Presentation Text Mining (33)

We start with an overview of the different steps (data analytics) in web search. This is followed by Link Structure Analysis including Hubs, Authorities and PageRank. The application of PageRank ideas as reputation outside web search is covered. Issues in web advertising and search follow. his leads to emerging field of computational advertising. The use of clustering and topic models completes unit with Google News as an example.

Video Web Search and Text Mining II (6:11)

This short lesson describes the different steps needed in web search including: Get the digital data (from web or from scanning); Crawl web; Preprocess data to get searchable things (words, positions); Form Inverted Index mapping words to documents; Rank relevance of documents with potentially sophisticated techniques; and integrate technology to support advertising and ways to allow or stop pages artificially enhancing relevance.

Video Related Applications (17:24)

The value of links and the concepts of Hubs and Authorities are discussed. This leads to definition of PageRank with examples. Extensions of PageRank viewed as a reputation are discussed with journal rankings and university department rankings as examples. There are many extension of these ideas which are not discussed here although topic models are covered briefly in a later lesson.

Video Web Advertising and Search (9:02)

Internet and mobile advertising is growing fast and can be personalized more than for traditional media. There are several advertising types Sponsored search, Contextual ads, Display ads and different models: Cost per viewing, cost per clicking and cost per action. This leads to emerging field of computational advertising.

Clustering and Topic Models

Video Clustering and Topic Models (6:21)

We discuss briefly approaches to defining groups of documents. We illustrate this for Google News and give an example that this can give different answers from word-based analyses. We mention some work at Indiana University on a Latent Semantic Indexing model.

Resources

All resources accessed March 2018.

14 - WebPlotViz

WebPlotViz

WebPlotViz is a browser based visualization tool developed at Indiana University. This tool allows user to visualize 2D and 3D data points in the web browser. WebPlotViz was developed as a succesor to the previous visualization tool PlotViz which was a application which needed to be installed on your machine to be used. You can find more information about PlotViz at the PlotViz Section

Motivation

The motivation of WebPlotViz is similar to PlotViz which is that the human eye is very good at pattern recognition and can see structure in data. Although most Big data is higher dimensional than 3, all data can be transformed by dimension reduction techniques to 3D and one can check analysis like clustering and/or see structure missed in a computer analysis.

How to use

In order to use WebPlotViz you need to host the application as a server, this can be done on you local machine or a application server. The source code for WebPlotViz can be found at the git hub repo WebPlotViz git Repo.

However there is a online version that is hosted on Indiana university servers that you can access and use. The online version is available at WebPlotViz

In order to use the services of WebPlotViz you would need to first create a simple account by providing your email and a password. Once the account is created you can login and upload files to WebPlotViz to be visualized.

Uploading files to WebPlotViz

While WebPlotViz does accept several file formats as inputs, we will look at the most simple and easy to use format that users can use. Files are uploaded as “.txt” files with the following structure. Each value is separated by a space.

Index x_val y_val z_val cluster_id label

Example file:

0 0.155117377 0.011486086 -0.078151964 1 l1
1 0.148366394 0.010782429 -0.076370584 2 l2
2 0.170597667 -0.025115137 -0.082946074 2 l2
3 0.136063907 -0.006670781 -0.082583441 3 l3
4 0.158259943 0.015187686 -0.073592601 5 l5
5 0.162483279 0.014387166 -0.085987414 5 l5
6 0.138651632 0.013358333 -0.062633719 5 l5
7 0.168020213 0.010742307 -0.090281011 5 l5
8 0.15810229 0.007551404 -0.083311109 4 l4
9 0.146878082 0.003858649 -0.071298345 4 l4
10 0.151487542 0.011896318 -0.074281645 4 l4

Once you have the data file properly formatted you can upload the file through the WebPlotViz GUI. Once you login to your account you should see a Green “Upload” button on the top left corner. Once you press it you would see a form that would allow you to choose the file, provide a description and select a group to which the file needs to be categorized into. If you do not want to assign a group you can simply use the default group which is picked by default

Once you have uploaded the file the file should appear in the list of plots under the heading “Artifacts”. Then you can click on the name or the “View” link to view the plot. Clicking on “View” will directly take you to the full view of the plot while clicking on the name will show and summary of the plot with a smaller view of the plot (Plot controls are not available in the smaller view). You can view how the sample dataset looks like after uploading at the following link. @fig:webpviz-11 shows a screen shot of the plot.

11 Points WebPlotViz plot

11 Points Plot{#fig:webpviz-11}

Users can apply colors to clusters manually or choose one of the color schemes that are provided. All the controls for the clusters are made available once your clock on the “Cluster List” button that is located on the bottom left corner of the plot (Third button from the left). This will pop up a window that will allow you to control all the settings of the clusters.

Features

WebPlotViz has many features that allows the users to control and customize the plots, Other than simple 2D/3D plots, WebPlotViz also supports time series plots and Tree structures. The examples section will show case examples for each of these. The data formats required for these plots are not covered here.

WebPlotViz Features labeled{#fig:webpviz-labled}

Some of the features are labeled in @fig:webpviz-labled. Please note that @fig:webpviz-labled shows an time series plot so the controls for playback shown in the figure are not available in single plots.

Some of the features are descibed in the short video that is linked in the home page of the hosted WebPlotViz site WebPlotViz

Examples

Now we will take a look at a couple of examples that were visualized using WebPlotViz.

Fungi gene sequence clustering example

The following example is a plot from clustering done on a set on fungi gene sequence data.

Fungi Gene Sequence Plot

Fungi Gene Sequence{#fig:webpviz-fungi}

Stock market time series data

This example shows a time series plot, the plot were created from stock market data so certain patterns can be followed with companies with passing years.

Stock market data

Stock market data{#fig:webpviz-stock}