Open Source Solutions

Over the past few months, a debate has been proceeding on whether or not a new discipline, a new career path, is emerging from the tsunami of data bearing down on us. The need for a new type of Renaissance [Wo]Man to deal with the Big Data onslaught. To whit, Data Science.

I'm writing about this now, because last night, at an every-three-week get together devoted to cask beer and data analysis, the topic came up. [Yes, every-THREE-weeks - a month is too long to go without cask beer fueled discussions of Rstats, BigData, Streaming SQL, BI and more.] The statisticians in the group, including myself, strongly disagreed with the way the term is being used; the software/database types were either in favor or ambivalent. We all agreed that a new, interdisciplinary approach to Big Data is needed. Oh, and I'll stay on topic here, and not get into another debate as to the definition of "Big Data".

This lively conversation reinforced my desire to write about Data Science that swelled up in me after reading "What is Data Science?" by Mike Loukides published on O'Reilly Radar, and a subsequent discussion on Twitter held the following weekend, concerning data analytics.

The term "Data Science" isn't new, but it is taking on new meanings. The Journal of Data Science published JDS volume 1, issue 1 in January of 2003. The Scope of the JDS is very clearly related to applied statistics

By "Data Science", we mean almost everything that has something to do with data: Collecting, analyzing, modeling...... yet the most important part is its applications --- all sorts of applications. This journal is devoted to applications of statistical methods at large.
-- About JDS, Scope, First Paragraph

There is also the CODATA Data Science Journal, which appears to have last been updated in August of 2007, and currently has no content, other than its self-description as

The Data Science Journal is a peer-reviewed electronic journal publishing papers on the management of data and databases in Science and Technology.

I think that two definitions can be derived from these two journals.

1. Data Science is systematic study, through observation and experiment, of the collection, modeling, analysis, visualization, dissemination, and application of data.
2. Data Science is the use of data and database technology within physical and natural sciences and engineering.

I can agree with the first, especially with the JDS Scope clearly stating that Data Science is applied statistics.

The New Oxford American Dictionary, on which the Apple Dictionary program is based, defines science as a noun

the intellectual and practical activity encompassing the systematic study of the structure and behaviour of the physical and natural world through observations and experiments.

And a similar definition of science can be found on Dictionary.com.

In many ways, I like Mike Loukides' article "What is Data Science?" in how it highlights the need for this new discipline. I just don't like what he describes to be the new definition of "data science". Indeed, I very much disagree with this statement from the article.

Using data effectively requires something different from traditional statistics, where actuaries in business suits perform arcane but fairly well-defined kinds of analysis. What differentiates data science from statistics is that data science is a holistic approach. We're increasingly finding data in the wild, and data scientists are involved with gathering data, massaging it into a tractable form, making it tell its story, and presenting that story to others.

A statistician is not an actuary. They're very different roles. I know this because I worked for over a decade applying statistics to determining the reliability and risk associated with very large, complex systems such as rockets and space-borne astrophysics observatories. I once hired a Cal student as an intern because she feared that the only career open to her as a math major, was to be an actuary. I showed her a different path. So, yes, I know, from experience, that a statistician is not an actuary. Actually, the definition of a data scientist given, that is "gathering data, massaging it into a tractable form, making it tell its story, and presenting that story to others" is exactly what a statistician does.

I do however see the need for a new discipline, separate from applied statistics, or data science. The massive amount of data to come from an instrumented world with strongly interconnected people and machines, and real-time analysis, inference and prediction from those data, will require inter-disciplinary skills. But I see those skills coming together in a person who is more of a smith, or, as Julian Hyde put it last night, an artisan. Falling back on the old dictionary again, a smith is someone who is skilled in creating something with a specific material; an artisan is someone who is skilled in a craft, making things by hand.

Another reason that I don't like the term "data science" for this interdisciplinary role, stems from what Mike Loukides describes in his article "What is Data Science?" as the definition for this new discipline "Data science requires skills ranging from traditional computer science to mathematics to art". I agree that the new discipline requires these three things, and more, even softer skills. I disagree that these add up to data science.

I even prefer "data geek", as defined by Michael E. Driscoll in "The Three Sexy Skills of Data Geeks". Michael Driscoll's post of 2009 May 27 certainly agrees skill-wise with Mike Loukides post of 2010 June 02.

1. Skill #1: Statistics (Studying)
2. Skill #2: Data Munging (Suffering)
3. Skill #3: Visualization (Storytelling)

And I very much prefer "Data Munging" to "Computer Science" as one of the three skills.

I'll stick to the definition that I gave above for data science as "systematic study, through observation and experiment, of the collection, modeling, analysis, visualization, dissemination, and application of data". This is also applied statistics. So, what else is needed for this new discipline? Well, Mike and Michael are correct: computer skills, especially data munging, and art. Well, any statistician today has computer skills, generally in one or more of SAS, SPSS, R, S-plus, Python, SQL, Stata, MatLab and other software packages, as well as familiarity with various data storage & management methods. Some statisticians are even artists, perhaps as story tellers, as evidenced by that rare great teacher or convincing expert witness, perhaps as visualizers, creating statistically accurate animations to clearly describe the analysis, as evidenced by the career of that intern I hired so many years ago.

The data smith, the data artisan, must be comfortable with all forms of data:

• structured,
• unstructured and
• semi-structured

Just as any other smith, someone following this new discipline might serve an apprenticeship creating new things from these forms of data such as a data warehouse or an OLAP cube, a sentiment analysis or a streaming SQL sensor web, or a recommendation engine or complex system predictives. The data smith must become very comfortable with putting all forms of data together in new ways, to come to new conclusions.

Just as a goldsmith will never make a piece of jewelry identical to the one finished days before, just as art can be forged but not duplicated, the data smith, the data artisan will glean new inferences every time they look at the data, will make new predictions with every new datum, and the story they tell, the picture they paint, will be different each time.

And perhaps then, the data smith becomes a master, an artisan.

Except where otherwise noted, this content is
licensed under a Creative Commons License.

PS: Here's a list of links to that Twitter conversation among some of the most respected people in the biz, on Data Analytics

1. https://twitter.com/NeilRaden/status/15512935981
2. https://twitter.com/NeilRaden/status/15513225191
3. https://twitter.com/NeilRaden/status/15513275261
4. https://twitter.com/NeilRaden/status/15513453916
5. https://twitter.com/datachick/status/15513460384
6. https://twitter.com/NeilRaden/status/15513488053
7. https://twitter.com/datachick/status/15513677836
8. https://twitter.com/CMastication/status/15513772446
9. https://twitter.com/NeilRaden/status/15513821393
10. https://twitter.com/NeilRaden/status/15513854916
11. https://twitter.com/NeilRaden/status/15513915694
12. https://twitter.com/alecsharp/status/15513980301
13. https://twitter.com/NeilRaden/status/15514104372
14. https://twitter.com/alecsharp/status/15514097194
15. https://twitter.com/CMastication/status/15514374095
16. https://twitter.com/estrenuo/status/15514634644
17. https://twitter.com/NeilRaden/status/15515243453
18. https://twitter.com/CMastication/status/15516185085
19. https://twitter.com/annmariastat/status/15516321715
20. https://twitter.com/NeilRaden/status/15519544709
21. https://twitter.com/NeilRaden/status/15519597061
22. https://twitter.com/NeilRaden/status/15519621974
23. https://twitter.com/skemsley/status/15519932631
24. https://twitter.com/skemsley/status/15519932631
25. https://twitter.com/aristippus303/status/15520146540
26. https://twitter.com/NeilRaden/status/15520478566
27. https://twitter.com/SethGrimes/status/15520765766
28. https://twitter.com/SethGrimes/status/15520851678
29. https://twitter.com/NeilRaden/status/15521050387
30. https://twitter.com/NeilRaden/status/15521106901
31. https://twitter.com/NeilRaden/status/15521133647
32. https://twitter.com/NeilRaden/status/15521192977
33. https://twitter.com/SethGrimes/status/15521579977
34. https://twitter.com/ryanprociuk/status/15521637974

'Tis been longer than intended, but we finally have the technology, time and resources to continue with our Open Source Solutions Decision Support System Study Guide (OSS DSS SG).

First, I want to thank SQLstream for allowing us to use SQLstream as a part of our solution. As mentioned in our "First DSS Study Guide" post, we were hoping to add a real-time component to our DSS. SQLstream is not open source, and not readily available for download. It is however, a co-founder and core contributer to the open source Eigenbase Project, and has incorporated Eigenbase technology into its product. So, what is SQLstream? To quote their web site, "SQLstream enables executives to make strategic decisions based on current data, in flight, from multiple, diverse sources". And that is why we are so interested in having SQLstream as a part of our DSS technology stack: to have the capability to capture and manipulate data as it is being generated.

Today, there are two very important classes of technologies that should belong to any DSS: data warehousing (DW) and business intelligence (BI). What actually comprises these technologies is still a matter of debate. To me, they are quite interrelated and provide the following capabilities.

• The means of getting data from one or more sources to one or more target storage & analysis systems. Regardless of the details for the source(s) and the target(s), the traditional means in data warehousing is Extract from the source(s), Transform for consistency & correctness, and Load into the target(s), that is, ETL. Other means, such as using data services within a services oriented architecture (SOA) either using provider-consumer contracts & Web Service Definition Language (WSDL) or representational state transfer (ReST) are also possible.
• Active storage over the long term of historic and near-current data. Active storage as opposed to static storage, such as a tape archive. This storage should be optimized for reporting and analysis through both its logical and physical data models, and through the database architecture and technologies implemented. Today we're seeing an amazing surge of data storage and management innovation, with column-store relational database management systems (RDBMS), map-reduce (M-R), key-value stores (KVS) and more, especially hybrids of one or several of old and new technologies. The innovation is coming so thick and fast, that the terminology is even more confused than in the rest of the BI world. NoSQL has become a popular term for all non-RDBMS, and even some RDBMS like column-store. But even here, what once meant No Structured Query Language now is often defined as Not only Structured Query Language, as if SQL was the only way to create an RDBMS (can someone say Progress and its proprietary 4GL).
• Tools for reporting including gathering the data, performing calculations, graphing, or perhaps more accurately, charting, formating and disseminating.
• Online Analytical Processing (OLAP) also known as "slice and dice", generally allowing forms of multi-dimensional or pivot analysis. Simply put, there are three underlying concepts for OLAP: the cube (a.k.a. hypercube, multi-dimensional database [MDDB] or OLAP engine), the measures (facts) & dimensions, and aggregation. OLAP provides much more flexibility than reporting, though the two often work hand-in-hand, especially for ad-hoc reporting and analysis.
• Data Mining, including machine learning and the ability to discover correlations among disparate data sets.

For our purposes, an important question is whether or not there are open source, or at least open source based, solutions for all of these capabilities. The answer is yes. As a matter of fact, there are three complete open source BI Suites [there were four, but the first, written in PERL, the Bee Project from the Czech Republic, is no longer being updated]. Here's a brief overview of SpagoBI, JasperSoft, and Pentaho.

We'll be using Pentaho, but you can use any of the these, or any combination of the OSS projects that are used by these BI Suites, or pick and choose from the more than 60 projects in our OSS Linkblog, as shown in the sidebar to this blog. All of the OSS BI Suites have many more features than shown in the simple table above. For example, SpagoBI has good tools for geographic & location services. Also, JasperSoft Professional and Enterprise Editions have many features than their Community Edition, such as Ad Hoc Reporting and Dashboards. Pentaho has a different Analyzer in their Enterprise Edition than either jPivot or PAT, Pentaho Analyzer, based upon the SaaS ClearView from the now-defunct LucidEra, as well as ease-of-use tools such as an OLAP schæma designer, and enterprise class security and administration tools.

Data warehousing using general purpose RDBMS systems such as Oracle, EnterpriseDB, PostrgeSQL or MySQL, are gradually giving way to analytic database management system (ADBMS), or, as we mentioned above, the catch-all NoSQL data storage systems, or even hybrid systems. For example, Oracle recently introduced hybrid column-row store features, and Aster Data has a column-store Massive Parallel Processing (MPP) DBMS|map-reduce hybrid [updated 20100616 per comment from Seth Grimes]. Pentaho supports Hadoop, as well as traditional general purpose RDBMS and column-store ADMBS. In the open source world, there are two columnar storage engines for MySQL, Infobright and Calpont InfiniDB, as well as one column-store ADBMS purpose built for BI, LucidDB. We'll be using LucidDB, and just for fun, may throw some data into Hadoop.

In addition, a modern DSS needs two more primary capabilities. Predictives, sometimes called predictive intelligence or predictive analytics (PA), which is the ability to go beyond inference and trend analysis, assigning a probability, with associated confidence, or likelihood of an event occurring in the future, and full Statistical Analysis, which includes determining the probability density or distribution function that best describes the data. Of course, there are OSS projects for these as well, such as The R Project, the Apache Common Math libraries, and other GNU projects that can be found in our Linkblog.

For statistical analysis and predictives, we'll be using the open source R statistical language and the open standard predictive model markup language (PMML), both of which are also supported by Pentaho.

We have all of these OSS projects installed on a Red Hat Enterprise Linux machine. The trick will be to get them all working together. The magic will be in modeling and analyzing the data to support good decisions. There are several areas of decision making that we're considering as examples. One is fairly prosaic, one is very interesting and far-reaching, and the others are somewhat in between.

1. A fairly simple example would be to take our blog statistics, a real-time stream using SQLstream's Twitter API, and run experiments to determine whether or not, and possibly how, Twitter affects traffic to and interaction with our blogs. Possibly, we could get to the point where we can predict how our use of Twitter will affect our blog.
2. A much more far-reaching idea was presented by Ken Winnick to me, via Twitter, and has created an on-going Twitter conversation and hashtag, #BPgulfDB. Let's take crowd sourced, government, and other publicly available data about the recent oilspill in the Gulf of Mexico, and analyze it.
3. Another idea is to take historical home utility usage plus current smart meter usage data, and create a real-time dashboard, and even predictives, for reducing and managing energy usage.
4. We also have the opportunity of using public data to enhance reporting and analytics for small, rural and research hospitals.

Except where otherwise noted, this content is
licensed under a Creative Commons License.

The next step in our open source solutions (OSS) for decision support systems (DSS) study guide (SG), according to the syllabus, is to make our first decision: a formal definition of "Decision Support System". Next, and soon, will be a post listing the technologies that will contribute to our studies.

The first stop in looking for a definition of anything today, is Wikipedia. And indeed, Wikipedia does have a nice article on DSS. One of the things that I find most informative about Wikipedia articles, is the "Talk" page for an article. The DSS discussion is rather mild though, no ongoing debate as can be found on some other talk pages, such as the discussion about Business Intelligence. The talk pages also change more often, and provide insight into the thoughts that go into the main article.

And of course, the second stop is a Google search for Decision Support System; a search on DSS is not nearly as fruitful for our purposes.

Once upon a time, we might have gone to a library and thumbed through the card catalog to find some books on Decision Support Systems. A more popular approach today would be to search Amazon for Decision Support books. There are several books in my library that you might find interesting for different reasons:

1. Pentaho Solutions: Business Intelligence and Data Warehousing with Pentaho and MySQL by Roland Bouman & Jos van Dongen provides a very good overview of data warehousing, business intelligence and data mining, all key components to a DSS, and does so within the context of the open source Pentaho suite
2. Smart Enough Systems: How to Deliver Competitive Advantage by Automating Hidden Decisions by James Taylor & Neil Raden introduces business concepts for truly managing information and using decision support systems, as well as being a primer on data warehousing and business intelligence, but goes beyond this by automating the data flow and decision making processes
3. Business Intelligence Roadmap: The Complete Project Lifecycle for Decision-Support Applications by Larissa T. Moss & Shaku Atre takes a business, program and project management approach to implementing DSS within a company, introducing fundamental concepts in a clear, though simplistic level
4. Competing on Analytics: The New Science of Winning by Thomas H. Davenport & Jeanne G. Harris in many ways goes into the next generation of decision support by showing how data, statistical and quantitative analysis within a context specific processes, gives businesses a strong lead over their competition, albeit, it does so at a very simplistic, formulaic level

These books range from being technology focused to being general business books, but they all provide insight into how various components of DSS fit into a business, and different approaches to implementing them. None of them actually provide a complete DSS, and only the first focuses on OSS. If you followed the Amazon search link given previously, you might also have noticed that there are books that show Excel as a DSS, and there is a preponderance of books that focus on the biomedical/pharmaceutical/healthcare industry. Another focus area is in using geographic information systems (actually one of the first uses for multi-dimensional databases) for decision support. There are several books in this search that look good, but haven't made it into my library as yet. I would love to hear your recommendations (perhaps in the comments).

From all of this, and our experiences in implementing various DW, BI and DSS programs, I'm going to give a definition of DSS. From a previous post in this DSS SG, we have the following:

A DSS is a set of processes and technology that help an individual to make a better decision than they could without the DSS.
-- Questions and Commonality

As we stated, this is vague and generic. Now that we've done some reading, let's see if we can do better.

A DSS assists an individual in reaching the best possible conclusion, resolution or course of action in stand-alone, iterative or interdependent situations, by using historical and current structured and unstructured data, collaboration with colleagues, and personal knowledge to predict the outcome or infer the consequences.

I like that definition, but your comments will help to refine it.

Note that we make no mention of specific processes, nor any technology whatsoever. It reflects my bias that decisions are made by individuals not groups (electoral systems not withstanding). To be true to our "TeleInterActive Lifestyle" I should point out that the DSS must be available when and where the individual needs to make the decision.

Any comments?

Except where otherwise noted, this content is
licensed under a Creative Commons License.

Recently, AnnMaria De Mars, PhD (multiple) and Dr. Peter Flom, PhD have stirred up a bit of a tempest in a tweet-pot, as well as in the statistical blogosphere, with comparisons of R and SAS, IBM/SPSS and the like. I've commented on both of their blogs, but decided to expand a bit here, as the choice of R is something that we planned to cover in a later post to our Open Source Solutions Decision Support Systems Study Guide. First, let me say that Dr. De Mars and Dr. Flom appear to have posted completely independently of each other, and further, that their posts have different goals.

In The Next Big Thing, Dr. De Mars is looking for the next big thing, both to keep her own career on-track, and to guide students into areas of study that will be survive in the job market in the coming decades. This is always difficult for mentors, as we can't always anticipate the "black swan" events that might change things drastically. The tempestuous nature of her post came from one little sentence:

Contrary to what some people seem to think, R is definitely not the next big thing, either. -- AnnMaria De Mars, The Next Big Thing, AnnMaria's Blog

In SAS vs. R, Introduction and Request, Dr. Flom starts a series comparing R and SAS from the standpoint of a statistician deciding upon tools to use.

There are several threads in Dr. De Mars post. I agree with Dr. De Mars that two of the "next big things" in data management & analysis are data visualization and dealing with unstructured data. I'm of the opinion that there is a third area, related to the "Internet of Things" and the tsunami of data that will be generated by it. These are conceptual areas, however. Dr. De Mars quickly moves on to discussing the tools that might be a part of the solutions of these next big things. The concepts cited are neither software packages nor computing languages. The software packages SAS, IBM/SPSS, Stata, Pentaho and the like, and the computing language S, with its open source distribution R, and its proprietary distribution S+ are none likely to be the next big things, as they are currently useful tools to know.

I find it interesting that both Dr. De Mars and Dr. Flom, as well as the various commenters, tweeters, and other posters, are comparing software suites and applications with a computing language. I think that a bit more historical perspective might be needed in bringing these threads together.

In 1979, when I first sat down with a FORTRAN programmer to turn my Bayesian methodologies into practical applications to determine the reliability and risk associated with the STAR48 kick motor and associated Payload Assist Module (PAM), the statistical libraries for FORTRAN seemed amazing. The ease with which we were able to create the program and churn through decades of NASA data (after buying a 1MB memory box for the mainframe) was wondrous

Today, there's not so much wonder from such a feat. The evolution of computing has drastically affected the way in which we apply mathematics and statistics today. Several of the comments to these posts argue both sides of the statement that anyone doing statistics today should be a programmer, or shouldn't. It's an interesting argument, that I've also seen reflected in chemistry, as fewer technicians are used in the lab, and the Ph.D.s work directly with the robots to prepare the samples and interpret the results.

Approximately 15 years ago, I moved from solving scientific and engineering problems directly with statistics, to solving business problems through vendor's software suites. The marketing names for this endeavor have gone through several changes: Decision Support Systems, Very Large Databases, Data Warehousing, Data Marts, Corporate Information Factory, Business Intelligence, and the like. Today, Data Mining, Data Visualization, Sentiment Analysis, "Big Data", SQL Streaming, and similar buzzwords reflect the new "big thing". Software applications, from new as well as established vendors, both open source and proprietary, are coming to the fore to handle these new areas that represent real problems.

So, one question to answer for students, is which, if any, of these software packages will best survive with and aid the growth of, their maturing careers. Will Tableau, LyzaSoft, QlikView or Viney@rd be in a better spot in 20 years, through growth or acquisition, than SAS or IBM/SPSS? Will the open source movement take down the proprietary vendors or be subsumed by them? Is Pentaho/Weka the BI & data mining solution for their career? Maybe, maybe not. But what about that other beast of which everyone speaks? Namely, R, the r-project, the R Statistical Language. What is it? Is it a worthy alternative to SAS or IBM/SPSS or Pentaho/Weka? Or is it a different genus altogether? That's a question I've been seeking to answer for myself, in my own career evolution. After 15 years, software such as SAP/Business Objects and IBM/Cognos, haven't evolved into anything that I like, with their pinnacle of statistical computation being the "average", the arithmetic mean. SAS and IBM/SPSS are certainly better, and with data mining, machine learning and predictives becoming important to business, certainly likely to be a good choice for the future. But are they really powerful enough? Are they flexible enough? Can they be used to solve the next generation of data problems?  They're very likely to evolve into software that can do so.  But how quickly?  And like all vendor software, they have limitations based upon the market studies and business decisions of the corporation.

How is R different?

Well, first, R is a computing language. Unlike SAP/Business Objects, IBM/Cognos, IBM/SPSS, SAS, Pentaho, JasperSoft, SpagoBI, or Oracle, it's not a company, nor a BI Suite, nor even a collection of software applications.  Second, R is an open source project. It's an open source implementation of S. Like C, and the other single letter named languages, S came out of Bell Labs, and in the case of R, in 1976. The open source implementation, R comes from R. Ihaka and R. Gentleman, first revealed in 1996 through the article, R: A language for data analysis and graphics. Journal of Computational and Graphical Statistics, 5:299–314, and is often associated with the Department of Statistics, University of Auckland.

While I'm not a software engineer, R is a very compelling statistical tool. As a language, it's very intuitive… for a statistician. It's an interactive, interpretive, functional, object oriented, statistical programming language. R itself is written in R, C, C++ and FORTRAN; it's powerful. As an open source project, it has attracted thousands upon thousands of users who have formed a strong community. There are thousands upon thousands of community contributed packages for R. It's flexible, and growing. One of the main goals of R was data visualization, and it has a wonderful new package for data visualization in ggplot2. It's ahead of the curve. There are packages for  parallel processing (some quite specific), for big data beyond in-memory capacity, for servers, and for embedding in a web site.  Get the idea?  If you think you need something in R, search CRAN, RForge, BioConductor or Omegahat.

As you can tell, I like R. However, in all honesty, I don't think that the SAS vs. R controversy is an either/or situation. SAS, IBM/SPSS and Pentaho complement R and vice-versa. Pentaho, IBM/SPSS and some SAS products support R. R can read data from SAS, IBM/SPSS , relational databases, Excel, mapReduce and more. The real question isn't is one tool better than another but rather selecting the best tool to answer a particular question.  That being said, I'm looking forward to Dr. Flom's comparison, as well as the continuing discussion on Dr. De Mars' blog.

For us, the question is building a decision support system or stack from open source components. It looks like we'll have a good time doing so.

Except where otherwise noted, this content is
licensed under a Creative Commons License.

First, let me say that we're talking about systems supporting the decisions that are made by human beings, not "expert systems" that automate decisions.  As an example, let's look at inventory management.  A human might use various components of a DSS to determine the amount of an item in stock, the demand for that item as a trend to determine when it might be out of stock, and predictives as to various factors (internal, external, environmental, political, etc) that might affect supply, to come to a decision as to how much and when to order more of that item.  An expert system might be created that could also determine when and how much of an item to oder, using neural networks, Bayesian nets or other algorithms.  The expert system might even take from the same DSS components (or directly from their underlying data) as the human might.  One could even run the expert system in parallel with humans making the decisions, scoring or otherwise evaluating the two, until the expert system is comparable or better than the expert system.  But, we're not really interested in expert systems in this study guide.  We'll be focusing on systems that help humans to make better decisions, not on automated feedback and control loops.

To me, a technology doesn't matter very much if it's not supporting some process, or a step within a process.  That process may be for personal reasons or supporting work activities. For this study guide, let's begin by continuing the discussion that we began in the previous posts, about the process by which one makes a decision, the steps, the events, the triggers and the consequences of making a decision.

I have my own process in making decisions.  I've played in executive and management roles for many years, and have been responsible for 5 P/L centers.  But this is a study guide, and while I intend to offer my own opinions and interpretations, we need some objective sources to study.  Let's start with a Google search.  Of course, Wikipedia has an article.  A site of which I've not heard before has the first hit with their article on problem-solving and decision-making.  Science Daily has a timely article from 2010 March 13 on how we really make decisions, our brain activity during decision making.  I also like the map from The Institute for Strategic Clarity.  Mindtools sets out a list of techniques and tools for aiding in the decision making process, and provides an important caveat "Do remember, though, that the tools in this chapter exist only to assist your intelligence and common sense. These are your most important assets in good Decision Making".  Reading through various reviews, the one book on decision making that I want to add to my library is The Managerial Decision-Making Process, 5th ed. by E. Frank Harrison.  From the Glossary of Political Economy Terms, we have:

Where formal organizations are the setting in which decisions are made, the particular decisions or policies chosen by decision-makers can often be explained through reference to the organization's particular structure and procedural rules. Such explanations typically involve looking at the distribution of responsibilities among organizational sub-units, the activities of committees and ad hoc coordinating groups, meeting schedules, rules of order etc. The notion of fixed-in-advance standard operating procedures (SOPs) typically plays an important role in such explanations of individual decisions made. -- Organizational process models of decision-making

Let's revisit and expand upon the summary that we gave in the third post in this series.

1. As an individual faced with making a decision, I may want input from others, I may want consensus, but in the end, it is an individual decision, and I will bear the fruits of having made that decision.

2. I need to put the problem, and my decision making, into context.  I have a variety of resources at my disposal to do so:

   • historical data
   • current information
   • structured data from transactional systems, master data, metadata, data warehouse, and other possible sources
   • unstructured data from blogs, wikis, Zotero libraries, Evernote, searches, bookmarks and similar sources
   • email
   • non-electronic correspondence, notes and conversations
   • personal experience
   • the experience of others garnered through water cooler and hallway conversations, formal meetings, twitter, phone calls and the like
     
3. Now I need to understand all of these facts, opinions and conjecture at my disposal.  Part of this sifting all of it through my internal filters, using my "gut".  Part is using the various reporting and analytical tools at my disposal, and then filtering those through my gut.  And really, this and the next point will constitute the majority of this OSS DSS Study Guide - the tools we use.
4. As I contemplate the various decisions that I might make from all of this, I want to understand the consequences of each potential decision: might this decision lead to a better product, more profit, less profit, broader market penetration, higher reliability, or even an alternate universe.
5. As I make this decision, I'll want to collaborate with others.  Ideally, I'll want to collaborate within the context of my decision support system. Once upon a time we would do this by embedding the tools within a portal system, now we take a more master data management approach, and use a services oriented architecture with either web services description language (WSDL) or representational state transition (ReST) application programming interfaces (APIs) to the collaborative environment, usually a wiki.
   

In summary, this introduction has set up a framework for a decision-making process for an individual to use a decision support system.  The majority of this study guide will be to expore the actual decision support system, and the open source tools from which we can build such a system.

Except where otherwise noted, this content is
licensed under a Creative Commons License.

As promised, here's the syllabus for our study guide to decision support systems using open source solutions. We'll start with a first draft on 2010-03-23, and update and change based on ideas, comments and lessons learned. So, please comment. The updates will be marked. Deletions will be marked with a strike-though and not removed.

1. Introduction
   1. Continuing the discussion of the processes and technologies that constitute a decision support system
   2. Formalizing a definition of DSS as well as the components, such as business intelligence (BI) that contribute to a DSS
   3. Providing [and updating] the list of references for this study guide
2. Preparation
   1. Discussing the technology for use in this study guide including the client(s) and server (Red Hat Enterprise Linux 5)
   2. Checking for prerequisites for the open source solutions that will be used
   3. Hands-on exercises for preparing the system
3. Installation
   1. Pointers and examples for installing the open source server-side packages including but not limited to:
      1. LucidDB
      2. Pentaho BI-Server, including PAT, and Administrative Console
      3. RServe and/or RApache
   2. Pointers for installation of client-side software and some examples on MacOSX
4. Modeling
   1. Generally, we would determine the models, the architecture and then one (or more competing) design(s) to satisfy that architecture, including selecting the right technical solutions for the job at hand. Here, we're creating a learning environment for certain tools, so we're introducing the architecture and design studies after the technology installs.
   2. In general, this section will explore the various means of modeling processes, systems and data, specifically as these relate to making decisions.
   3. Decision Making Processes
      1. Decision Theory
      2. Game Theory
      3. Machine Learning & Data Mining
      4. Bayes and Iterations
      5. Predictives
   4. Information Flow
   5. Mathematical Modeling
   6. Data Modeling
   7. UML
   8. Dimensional Modeling
   9. PMML
5. Architecture and Design
   1. In this section, we'll examine the differences between enterprise and system architecture, and between architecture and design. We'll look at various architectural and design elements that might influence both policy and technology directions.
   2. Discussing Enterprise Architecture, especially the translation between the user needs and technology/operational realities
   3. System Architecture
   4. SOA, ReST, WSDL, and Master Data Management
   5. Technology selection and vendor bake-offs
6. Implementation Considerations
   1. Discussing the various philosophies and considerations for implementing any DSS, or really, any system integration project. We'll look at our own three track implementation methodology, as well as how the new Pentaho Agile BI tools support our method. In addition, we'll consider how we'll get all these OSS tools working together, on the same data sets, as well as, the importance of managing data about the data.
   2. Pentaho Agile BI and our own 8D™ Method
   3. System and Data Integration
   4. Metadata
7. Using the Tools
   1. This is the vaguest part of our syllabus. We'll be using the examples from our various references, but with the system we've set-up here, rather than the exact systems that the references use. For example, we'll be using LucidDB and not MySQL for the examples from Pentaho Solutions. Remember too, that this is a study guide, and not a oops meant to be a book written as a series of blog posts, so while we might vary from the reference materials, we'll always refer to them.
   2. ETL
   3. Reporting
   4. OLAP
   5. Data Mining & Machine Learning
   6. Statistical Analysis
   7. Predictives
   8. Workflow
   9. Collaboration
   10. Hmm, this should take years

Except where otherwise noted, this content is
licensed under a Creative Commons License.

In the introduction to our open source solutions (OSS) for decision support systems (DSS) study guide (SG), I gave a variety of examples of activities that might be considered using a DSS. I asked some questions as to what common elements exist among these activities that might help us to define a modern platform for DSS, and whether or not we could build such a system using open source solutions.

In this post, let's examine the first of those questions, and see if we can start answering those questions. In the next post, we will lay out a syllabus of sorts for this OSS DSS SG.

The first common element is that in all cases, we have an individual doing the activity, not a machine nor a committee.

Secondly, the individual has some resources at their disposal. Those resources include current and historical information, structured and unstructured data, communiqués and opinions, and some amount of personal experience, augmented by the experience of others.

Thirdly, though not explicit, there's the idea of digesting these resources and performing formal or informal analyses.

Fourthly, though again, not explicit, the concept of trying to predict what might happen next, or as a result of the decision is inherent to all of the examples.

Finally, there's collaboration involved. Few of us can make good decisions in a vacuum.

Of course, since the examples are fictional, and created by us, they represent our biases. If you had fingered our domain server back in 1993, or read our .project and .plan files from that time, you would have seen that we were interested in sharing information and analyses, while providing a framework for making decisions using such tools as email, gopher and electronic bulletin boards. So, if you identify any other commonalities, or think anything is missing, please join the discussion in the comments.

From these commonalities, can we begin to answer the first question we had asked: "What does this term [DSS] really mean?". Let's try.

A DSS is a set of processes and technology that help an individual to make a better decision than they could without the DSS.

That's nice and vague; generic enough to almost meaningless, but provides some key points that will help us to bound the specifics as we go along. For example, if a process or technology doesn't help us to make a better decision, than it doesn't fit. If something allows us to make a better decision, but we can't define the process or identify the technology involved, it doesn't belong (e.g. "my gut tells me so").

Let's create a list from all of the above.

1. Individual Decision Maker
2. Process
3. Technology
4. Structured Data
5. Unstructured Data
6. Historical Information
7. Current Information
8. Communication
9. Opinion
10. Collaboration
11. Analysis
12. Prediction
13. Personal Experience
14. Other's Experience

What do you think? Does a modern system to support decisions need to cover all of these elements and no others? Is this list complete and sufficient? The comments are open.

Someone sitting in their study, looking at their books, journals, piles of scholarly periodicals and files of correspondence with learned colleagues probably didn't think that they were looking at their decision support system, but they were.

Someone sitting on the plains, looking at the conditions around them, smoke signals from distant tribe members, records knotted into a string, probably didn't think that they were looking at their decision support system, but they were.

Someone at the nexus of a modern military command, control, communications, computing and intelligence system, probably didn't think that they were looking at their decision support system, but they were.

Someone pulling data from transactional systems, and dumping the results of reports & analyses from BI tool into a spreadsheet to feed a dashboard for the executives of a huge corporation probably didn't think that they were looking at their decision support system, but they were.

The term "decision support system" has been in use for over 50 years, perhaps longer.

• But what does this term really mean?
• What do all of my examples have in common?
• How can we build a reasonable decision support system from open source solutions?
• What resources exist to help us learn?

I'm starting a series of posts, essentially a "study guide" to help answer these questions.

I'll be drawing from and pointing to the following books and online resources as we install, configure and use open source systems to create a technical platform for a decision support system.

1. Bayesian Computation in R by Jim Albert, Springer Series in UseR!, ISBN: 0-38-792297-0, Purchase from Amazon, you can also purchase the Kindle ebook from Amazon
2. R in a Nutshell by Joseph Adler, ISBN: 0-59-68017-0X, Purchase from Amazon
3. Pentaho Solutions; Business Intelligence and Data Warehousing with Pentaho and MySQL, by Roland Bouman and Jos van Dongen, ISBN: 0-47-048432-2, Purchase from Amazon
4. Pentaho Reporting 3.5 for Java Developers by Will Gorman, ISBN: 1-84-719319-6, Purchase from Amazon
5. Pentaho Kettle Solutions: Building Open Source ETL Solutions with Pentaho Data Integration by Matt Casters, Roland Bouman & Jos van Dongen, ISBN: 0-47-063517-7 due 2010 September, Pre-Order from Amazon
6. Data Mining: Practical Machine Learning Tools and Techniques by Ian H. Witten and Eibe Frank, Second Edition, Morgan-Kaufmann Series in Data Management Systems, ISBN: 0-12-088407-0 a.k.a. "The Weka Book", Purchase from Amazon, Pre-Order the Third Edition, you can also purchase the Kindle ebook from Amazon
7. LucidDB online documentation
8. Pertinent information from Eigenbase
9. LudidDB mailing list archive on Nabble
10. Anything I can find on PAT
11. Pentaho Community Forums, Wiki, WebEx Events, and other community sources
12. R Mailing Lists and Forums
13. Various Books in PDF from The R Project
14. Information Management and Open Source Solution Blogs from our side-column linkblogs

In this study guide series of posts:

• I'll show how the datawarehousing (DW) and business intelligence (BI) can be extended to include all the elements held in common from my DSS examples.
• We'll examine the open source solutions Pentaho, R, Rserve, Rapache, LucidDB and possibly Map-Reduce & Key-value-stores, and the related open source projects, communities and companies in terms of how they can be used to create a DSS.
• I would like to add a collaboration tool to the mix, as we do in our implementation projects, possibly Mindtouch, a ReSTful Wiki Platform.
• We may add one non-open source package, SQLStream, that's built upon open source elements from Eigenbase. This will allow us to add a real-time component to our DSS.
• I'll give my own experience in installing these packages and getting them to work together, with pointers to the resources listed above.
• We'll explore sample and public data sets with the DSS environment we created, again with pointers to and help from the resources listed.

The purpose of this series of posts is a study guide, not an online book written as a blog. The goal is to help us to define a modern DSS and build it out of open source solutions, while using existing resources.

Please feel free to comment, especially if there is anything that you feel should be included beyond what I've outlined here.

Here's a personal perspective and a bit of a personal history regarding mathematical modeling and predictives.

The 1980s were an exciting time for mathematical modeling of complex systems. At the time, there were two basic types of modeling: deterministic and stochastic (probability or statistics models). Within stochastic modeling, traditional statistics vs. Bayesian statistics was a burgeoning battleground. Physical simulations (often based upon deterministic models) were giving way to computer simulations (often based upon stochastic models, especially Monte Carlo Simulations). Two theories were popularized during this time: catastrophe theory and chaos theory; ultimately though, both of these theories proved incapable of prediction - the hallmark of a good mathematical model. A different type of modeling technique, based upon relational algebra, was also moving from the theoretical work of Ted Codd, to the practical implementations at (the company now known as) Oracle: data modeling.

Mathematical models are attempts to understand the complex by making simplifying assumptions. They are always a balance between complexity and accuracy. One nice example of the evolution of a deterministic mathematical model can be found in the Ideal Gas Laws, starting with Boyle's Law to Charles' Law to Gay-Lussac's Law to Avogadro's Law, culminating in the Ideal Gas Law, which all of saw in high school chemistry: PV=nRT.

Mathematical models are used in pretty much all fields of endeavor: physical sciences, all types of engineering, behavioral studies, and business. In the 1970's, I used deterministic electrochemical models to understand and predict the behaviour of various chemical stoichiometry for fuel cells and photovoltaic cells. In the 1980's, I used Bayesian statistics, sometimes combined with Monte Carlo Simulations to predict the reliability and risk associated with complex aerospace, utility and other systems.

The most popular use of Bayesian statistics was to expand the a priori knowledge of a complex system with subjective opinions. Likely the most famous application of Bayesian Statistics, at the time I became involved with the branch, was the Rand Corporation's Delphi Method. There was actually a joke in the Aerospace Industry about the Delphi Method:

A team of Rand consultants went to Werner von Braun to seek the expert opinion of the engineers working on a new rocket motor. The consultants explained their Delphi Method thusly. Prior to the first static test of the new rocket motor, they would ask, separately, each of the five engineers working on the new design their opinion of the rocket's reliability. Their opinions would form the Bayesian a priori distribution. After the test, they would reveal the results of the first survey and the test results, and ask the five engineers, collectively, their new opinion of the rocket's reliability. This would form the Bayesian a posteriori, from which the rocket's reliability would be predicted. Doctor von Braun said that he could save them some time. He gathered his team of rocket engineers, and asked them if they thought that the new rocket motor would fail. Each answered, as did Doctor Von Braun, "no" in German. "There, you see, five nines reliability, as specified." declared the good Doctor to the Rand consultants, "No need for any further study on your part."

Yep, it's a side splitter.

I didn't like this method, and did things a bit differently. My method involved gathering all the data for similar test and production models, weighting each relevant engineering variable, creating the a priori, fitting with Weibull Analysis, designing the Bayesian mathematical conjugate, using a detailed post-mortem of the first and subsequent tests of the system being analyzed, updating and learning as we went, to finally predict the reliability and risk for the system. I first used this on the Star48 perigee kick motor, and went on to refine and use this method for:

• a variety of apogee and perigee kick motors
• several components of the Space Transportation System
• the Extreme Ultraviolet Explorer
• Gravity Probe-B
• a halogen lamp pizza oven
• a methodology for failure mode, effects and criticality analysis of the electrical grid
• and many more systems

I started to call this method "objective Bayes", but that name was already taken by a branch of Bayesian statistics that uses a non-informative a priori. Several of my projects resulted in software programs, all in FORTRAN. The first was used as a justification for a 1 MB [no, not a mistake] "box" [memory] for the corporate mainframe. NASA had sent us detailed data on over 4,000 solid propellant rocket motors. Talk about "big data". I had a lot of fun doing this into the 1990's.

The next paradigm shift, for me personally, was learning data modeling, and focusing on business processes rather than engineering systems. Spending time at Oracle, including Richard Barker and his computer aided system engineering methods, I felt right at home. Rather than Bayesian Statistics, I would be using relational algebra and calculus for deterministic mathematical models of the data for the business processes being stored in a relational database management system. I very quickly got involved in very large databases, decision support systems, data warehousing and business intelligence.

I was surprised, and, after 17 years, continue to be surprised, how few data modelers agree with the statement in the preceding paragraph. I'm surprised how few data modelers go beyond entity-relationship diagrams; how few know or care about relational algebra and relational calculus. I'm amazed how few people realize that the arithmetic average computed in most "analytic" systems is a fairly useless measure of the underlying data, for most systems. I'm amazed that BI and analytic systems are still deterministic, and always go with simplicity over accuracy.

But computer power continues to expand. Moore's Law still rules. We can do better now. Things that used to take powerful main frames or even supercomputers can be done on laptops now. We no longer need to settle for simplicity over accuracy.

More importantly, the R Statistical Language has matured. Literally thousands and thousands of mathematical, graphical and statistical packages have been added to the CRAN, Omegahat and BioConductor repositories. Even the New York Times has published pieces about R.

It's once again time to move from deterministic to stochastic models.

Over the next few weeks, I hope to post a series of "study guides" that will focus on setting up a web-based environment consolidating SQL and MDX based analytics, as expressed in Pentaho and LucidDB open source projects, with R, and possibly SQLStream. Updated 20100314 to correct links (typos). Thanks to Doug Moran of Pentaho for catching this.

There have been many articles as well on "Big Data". As I commented on Merv Adrian's blog post request for "Ideas for SF Big Data Summit":

One area of discussion, which may appear to be for the “newbies” but is actually a matter of some debate, would be the definition of “big data”.

It really isn’t about the amount of data (TB & PB & more) so much as it is about the volumetric flow and timeliness of the data streams.

It’s about how data management systems handle the various sources of data as well as the interweaving of those sources.

It means treating data management systems in the same way that we treat the Space Transportation System, as a very large, complex system.

-- Comment by Joseph A. di Paolantonio, February 1, 2010 at 4:09 pm

I believe this because there is a huge amount of data about to come down the pipe. I'm not talking about the Semantic Web or the pidly little petabytes of web log and click-through data. I'm talking about the instrumented world. Something that's been in the making for ten years, and more: RFID, SmartDust, ZigBee, and more wired and wireless sensors, monitors and devices that will become a part of everything, everywhere.

Let me just cite two examples from something that is coming, is hyped, but not yet standardized, even if solid attempts at definition are being made: the SmartGrid. First, consider the fact that utility companies are distributing and using smart meters to replace manually read mechanical meters at homes and businesses; this will result in thousands of data points per day as opposed to one per month PER METER. The second is EPRI's copper-riding robot, as explained in a recent Popular Science. Think of the petabytes of data that these two examples will generate monthly. [Order the Smart Grid Dictionary: First Edition on Amazon]

The desire, the need, to analyze and make inferences from this data will be great. The need to actually predict from this data will be even greater, and will be a necessary element of the coming SmartGrid, and in making the instrumented world a better world for all of humanity.

Now if only we can avoid the likes of Skynet and Archangel.

On Twitter today, Lance Walter asked me to go into the Ark Business with him, and Gareth Greenaway asked for entertainment. It must be a rainy Friday afternoon

I'm not sure about Lance's offer, but I did tell Gareth the following story, from tweet-start to tweet-end. This isn't word for word as I tweeted. 'Tis a bit expanded, but the tale is the same.

Once upon a time there was a young penguin named Tux. Tux decided to set off on a journey through IT Land. Now IT Land is a dangerous place, full of hackers fighting crackers, and ruled by those in the Ivory Tower and the acolytes of the Megaliths.

Along the way, the adventurous Tux met the Dolphin, the Elephant and the Beekeeper. They made a pact on the Lucid glyph to become a Dynamo of IT, bringing power to the datasmiths of the Land.

They met many Titans from the Megaliths on their Quest. The Beekeeper used the open source bees to open the scrum along the way, blocking the hookers with their sharp claws.

Some of the Titans were helpful, some, not so much.

The Dolphin was empowered by the Sun. But the Sun was consumed by a powerful Oracle. The Elephant, too, gained a powerful ally, and they do Enterprise against the Oracle. The band of the Quest was broken, and Tux was sad.

The Era of Lucid thought ended, but the Dynamo yet powers the Lucid Glyph, and Tux can rely on the Dynamo and the Beekeeper to predict a future clear of the Oracle.

And thus this quest ends, but another soon begins, where Tux will meet new friends and new foes. Will Beastie and the dæmons be allies? Will the Paladin in the Red Hat be stalwart?

Perhaps we'll find out at OSCON, for Gareth suggested that an assemblage of geeks would enjoy this story, and we'll see if OSCON thinks our tales worthy of a keynote slot in 2010.

Do you recognize all the characters in this tale? Maybe the links will help.

What say you, OSCON? Would these tales make a worthy Keynote?