Denial of Service Attacks against Hospitals and Emergency Rooms

One of the most frightening possibilities of cyber warfare is to use remotely deployed , or timed intrusion malware to disturb, distort, deny health care services.

Computer Virus Shuts Down Georgia Hospital

A doctor in an Emergency Room depends on critical information that may save lives if it is electronic and comes on time. However this electronic information can be distorted (which is more severe than deleting it)

The electronic system of a Hospital can also be overwhelmed. If there can be built Stuxnet worms on   nuclear centrifuge systems (like those by Siemens), then the widespread availability of health care systems means these can be reverse engineered for particularly vicious cyber worms.

An example of prime area for targeting is Veterans Administration for veterans of armed forces, but also cyber attacks against electronic health records.

Consider the following data points-

http://threatpost.com/en_us/blogs/dhs-warns-about-threat-mobile-devices-healthcare-051612

May 16, 2012, 9:03AM

DHS’s National Cybersecurity and Communications Integration Center (NCCIC) issued the unclassfied bulletin, “Attack Surface: Healthcare and Public Health Sector” on May 4. In it, DHS warns of a wide range of security risks, including that could expose patient data to malicious attackers, or make hospital networks and first responders subject to disruptive cyber attack

http://publicintelligence.net/nccic-medical-device-cyberattacks/

National Cybersecurity and Communications Integration Center Bulletin

The Healthcare and Public Health (HPH) sector is a multi-trillion dollar industry employing over 13 million personnel, including approximately five million first-responders with at least some emergency medical training, three million registered nurses, and more than 800,000 physicians.

(U) A significant portion of products used in patient care and management including diagnosis and treatment are Medical Devices (MD). These MDs are designed to monitor changes to a patient’s health and may be implanted or external. The Food and Drug Administration (FDA) regulates devices from design to sale and some aspects of the relationship between manufacturers and the MDs after sale. However, the FDA cannot regulate MD use or users, which includes how they are linked to or configured within networks. Typically, modern MDs are not designed to be accessed remotely; instead they are intended to be networked at their point of use. However, the flexibility and scalability of wireless networking makes wireless access a convenient option for organizations deploying MDs within their facilities. This robust sector has led the way with medical based technology options for both patient care and data handling.

(U) The expanded use of wireless technology on the enterprise network of medical facilities and the wireless utilization of MDs opens up both new opportunities and new vulnerabilities to patients and medical facilities. Since wireless MDs are now connected to Medical information technology (IT) networks, IT networks are now remotely accessible through the MD. This may be a desirable development, but the communications security of MDs to protect against theft of medical information and malicious intrusion is now becoming a major concern. In addition, many HPH organizations are leveraging mobile technologies to enhance operations. The storage capacity, fast computing speeds, ease of use, and portability render mobile devices an optimal solution.

(U) This Bulletin highlights how the portability and remote connectivity of MDs introduce additional risk into Medical IT networks and failure to implement a robust security program will impact the organization’s ability to protect patients and their medical information from intentional and unintentional loss or damage.

(U) According to Health and Human Services (HHS), a major concern to the Healthcare and Public Health (HPH) Sector is exploitation of potential vulnerabilities of medical devices on Medical IT networks (public, private and domestic). These vulnerabilities may result in possible risks to patient safety and theft or loss of medical information due to the inadequate incorporation of IT products, patient management products and medical devices onto Medical IT Networks. Misconfigured networks or poor security practices may increase the risk of compromised medical devices. HHS states there are four factors which further complicate security resilience within a medical organization.

1. (U) There are legacy medical devices deployed prior to enactment of the Medical Device Law in 1976, that are still in use today.

2. (U) Many newer devices have undergone rigorous FDA testing procedures and come equipped with design features which facilitate their safe incorporation onto Medical IT networks. However, these secure design features may not be implemented during the deployment phase due to complexity of the technology or the lack of knowledge about the capabilities. Because the technology is so new, there may not be an authoritative understanding of how to properly secure it, leaving open the possibilities for exploitation through zero-day vulnerabilities or insecure deployment configurations. In addition, new or robust features, such as custom applications, may also mean an increased amount of third party code development which may create vulnerabilities, if not evaluated properly. Prior to enactment of the law, the FDA required minimal testing before placing on the market. It is challenging to localize and mitigate threats within this group of legacy equipment.

3. (U) In an era of budgetary restraints, healthcare facilities frequently prioritize more traditional programs and operational considerations over network security.

4. (U) Because these medical devices may contain sensitive or privacy information, system owners may be reluctant to allow manufactures access for upgrades or updates. Failure to install updates lays a foundation for increasingly ineffective threat mitigation as time passes.

(U) Implantable Medical Devices (IMD): Some medical computing devices are designed to be implanted within the body to collect, store, analyze and then act on large amounts of information. These IMDs have incorporated network communications capabilities to increase their usefulness. Legacy implanted medical devices still in use today were manufactured when security was not yet a priority. Some of these devices have older proprietary operating systems that are not vulnerable to common malware and so are not supported by newer antivirus software. However, many are vulnerable to cyber attacks by a malicious actor who can take advantage of routine software update capabilities to gain access and, thereafter, manipulate the implant.

(U) During an August 2011 Black Hat conference, a security researcher demonstrated how an outside actor can shut off or alter the settings of an insulin pump without the user’s knowledge. The demonstration was given to show the audience that the pump’s cyber vulnerabilities could lead to severe consequences. The researcher that provided the demonstration is a diabetic and personally aware of the implications of this activity. The researcher also found that a malicious actor can eavesdrop on a continuous glucose monitor’s (CGM) transmission by using an oscilloscope, but device settings could not be reprogrammed. The researcher acknowledged that he was not able to completely assume remote control or modify the programming of the CGM, but he was able to disrupt and jam the device.

http://www.healthreformwatch.com/category/electronic-medical-records/

February 7, 2012

Since the data breach notification regulations by HHS went into effect in September 2009, 385 incidents affecting 500 or more individuals have been reported to HHS, according to its website.

http://www.darkdaily.com/cyber-attacks-against-internet-enabled-medical-devices-are-new-threat-to-clinical-pathology-laboratories-215#axzz1yPzItOFc

February 16 2011

One high-profile healthcare system that regularly experiences such attacks is the Veterans Administration (VA). For two years, the VA has been fighting a cyber battle against illegal and unwanted intrusions into their medical devices

 

http://www.mobiledia.com/news/120863.html

 DEC 16, 2011
Malware in a Georgia hospital’s computer system forced it to turn away patients, highlighting the problems and vulnerabilities of computerized systems.

The computer infection started to cause problems at the Gwinnett Medical Center last Wednesday and continued to spread, until the hospital was forced to send all non-emergency admissions to other hospitals.

More doctors and nurses than ever are using mobile devices in healthcare, and hospitals are making patient records computerized for easier, convenient access over piles of paperwork.

http://www.doctorsofusc.com/uscdocs/locations/lac-usc-medical-center

As one of the busiest public hospitals in the western United States, LAC+USC Medical Center records nearly 39,000 inpatient discharges, 150,000 emergency department visits, and 1 million ambulatory care visits each year.

http://www.healthreformwatch.com/category/electronic-medical-records/

If one jumbo jet crashed in the US each day for a week, we’d expect the FAA to shut down the industry until the problem was figured out. But in our health care system, roughly 250 people die each day due to preventable error

http://www.pcworld.com/article/142926/are_healthcare_organizations_under_cyberattack.html

Feb 28, 2008

“There is definitely an uptick in attacks,” says Dr. John Halamka, CIO at both Beth Israel Deaconess Medical Center and Harvard Medical School in the Boston area. “Privacy is the foundation of everything we do. We don’t want to be the TJX of healthcare.” TJX is the Framingham, Mass-based retailer which last year disclosed a massive data breach involving customer records.

Dr. Halamka, who this week announced a project in electronic health records as an online service to the 300 doctors in the Beth Israel Deaconess Physicians Organization,

Interview Dan Steinberg Founder Salford Systems

Here is an interview with Dan Steinberg, Founder and President of Salford Systems (http://www.salford-systems.com/ )

Ajay- Describe your journey from academia to technology entrepreneurship. What are the key milestones or turning points that you remember.

 Dan- When I was in graduate school studying econometrics at Harvard,  a number of distinguished professors at Harvard (and MIT) were actively involved in substantial real world activities.  Professors that I interacted with, or studied with, or whose software I used became involved in the creation of such companies as Sun Microsystems, Data Resources, Inc. or were heavily involved in business consulting through their own companies or other influential consultants.  Some not involved in private sector consulting took on substantial roles in government such as membership on the President’s Council of Economic Advisors. The atmosphere was one that encouraged free movement between academia and the private sector so the idea of forming a consulting and software company was quite natural and did not seem in any way inconsistent with being devoted to the advancement of science.

 Ajay- What are the latest products by Salford Systems? Any future product plans or modification to work on Big Data analytics, mobile computing and cloud computing.

 Dan- Our central set of data mining technologies are CART, MARS, TreeNet, RandomForests, and PRIM, and we have always maintained feature rich logistic regression and linear regression modules. In our latest release scheduled for January 2012 we will be including a new data mining approach to linear and logistic regression allowing for the rapid processing of massive numbers of predictors (e.g., one million columns), with powerful predictor selection and coefficient shrinkage. The new methods allow not only classic techniques such as ridge and lasso regression, but also sub-lasso model sizes. Clear tradeoff diagrams between model complexity (number of predictors) and predictive accuracy allow the modeler to select an ideal balance suitable for their requirements.

The new version of our data mining suite, Salford Predictive Modeler (SPM), also includes two important extensions to the boosted tree technology at the heart of TreeNet.  The first, Importance Sampled learning Ensembles (ISLE), is used for the compression of TreeNet tree ensembles. Starting with, say, a 1,000 tree ensemble, the ISLE compression might well reduce this down to 200 reweighted trees. Such compression will be valuable when models need to be executed in real time. The compression rate is always under the modeler’s control, meaning that if a deployed model may only contain, say, 30 trees, then the compression will deliver an optimal 30-tree weighted ensemble. Needless to say, compression of tree ensembles should be expected to be lossy and how much accuracy is lost when extreme compression is desired will vary from case to case. Prior to ISLE, practitioners have simply truncated the ensemble to the maximum allowable size.  The new methodology will substantially outperform truncation.

The second major advance is RULEFIT, a rule extraction engine that starts with a TreeNet model and decomposes it into the most interesting and predictive rules. RULEFIT is also a tree ensemble post-processor and offers the possibility of improving on the original TreeNet predictive performance. One can think of the rule extraction as an alternative way to explain and interpret an otherwise complex multi-tree model. The rules extracted are similar conceptually to the terminal nodes of a CART tree but the various rules will not refer to mutually exclusive regions of the data.

 Ajay- You have led teams that have won multiple data mining competitions. What are some of your favorite techniques or approaches to a data mining problem.

 Dan- We only enter competitions involving problems for which our technology is suitable, generally, classification and regression. In these areas, we are  partial to TreeNet because it is such a capable and robust learning machine. However, we always find great value in analyzing many aspects of a data set with CART, especially when we require a compact and easy to understand story about the data. CART is exceptionally well suited to the discovery of errors in data, often revealing errors created by the competition organizers themselves. More than once, our reports of data problems have been responsible for the competition organizer’s decision to issue a corrected version of the data and we have been the only group to discover the problem.

In general, tackling a data mining competition is no different than tackling any analytical challenge. You must start with a solid conceptual grasp of the problem and the actual objectives, and the nature and limitations of the data. Following that comes feature extraction, the selection of a modeling strategy (or strategies), and then extensive experimentation to learn what works best.

 Ajay- I know you have created your own software. But are there other software that you use or liked to use?

 Dan- For analytics we frequently test open source software to make sure that our tools will in fact deliver the superior performance we advertise. In general, if a problem clearly requires technology other than that offered by Salford, we advise clients to seek other consultants expert in that other technology.

 Ajay- Your software is installed at 3500 sites including 400 universities as per http://www.salford-systems.com/company/aboutus/index.html What is the key to managing and keeping so many customers happy?

 Dan- First, we have taken great pains to make our software reliable and we make every effort  to avoid problems related to bugs.  Our testing procedures are extensive and we have experts dedicated to stress-testing software . Second, our interface is designed to be natural, intuitive, and easy to use, so the challenges to the new user are minimized. Also, clear documentation, help files, and training videos round out how we allow the user to look after themselves. Should a client need to contact us we try to achieve 24-hour turn around on tech support issues and monitor all tech support activity to ensure timeliness, accuracy, and helpfulness of our responses. WebEx/GotoMeeting and other internet based contact permit real time interaction.

 Ajay- What do you do to relax and unwind?

 Dan- I am in the gym almost every day combining weight and cardio training. No matter how tired I am before the workout I always come out energized so locating a good gym during my extensive travels is a must. I am also actively learning Portuguese so I look to watch a Brazilian TV show or Portuguese dubbed movie when I have time; I almost never watch any form of video unless it is available in Portuguese.

 Biography-

http://www.salford-systems.com/blog/dan-steinberg.html

Dan Steinberg, President and Founder of Salford Systems, is a well-respected member of the statistics and econometrics communities. In 1992, he developed the first PC-based implementation of the original CART procedure, working in concert with Leo Breiman, Richard Olshen, Charles Stone and Jerome Friedman. In addition, he has provided consulting services on a number of biomedical and market research projects, which have sparked further innovations in the CART program and methodology.

Dr. Steinberg received his Ph.D. in Economics from Harvard University, and has given full day presentations on data mining for the American Marketing Association, the Direct Marketing Association and the American Statistical Association. After earning a PhD in Econometrics at Harvard Steinberg began his professional career as a Member of the Technical Staff at Bell Labs, Murray Hill, and then as Assistant Professor of Economics at the University of California, San Diego. A book he co-authored on Classification and Regression Trees was awarded the 1999 Nikkei Quality Control Literature Prize in Japan for excellence in statistical literature promoting the improvement of industrial quality control and management.

His consulting experience at Salford Systems has included complex modeling projects for major banks worldwide, including Citibank, Chase, American Express, Credit Suisse, and has included projects in Europe, Australia, New Zealand, Malaysia, Korea, Japan and Brazil. Steinberg led the teams that won first place awards in the KDDCup 2000, and the 2002 Duke/TeraData Churn modeling competition, and the teams that won awards in the PAKDD competitions of 2006 and 2007. He has published papers in economics, econometrics, computer science journals, and contributes actively to the ongoing research and development at Salford.

High Performance Analytics

Marry Big Data Analytics to High Performance Computing, and you get the buzzword of this season- High Performance Analytics.

It basically consists of Parallelized code to run in parallel on custom hardware, in -database analytics for speed, and cloud computing /high performance computing environments. On an operational level, it consists of software (as in analytics) partnering with software (as in databases, Map reduce, Hadoop) plus some hardware (HP or IBM mostly). It is considered a high margin , highly profitable, business with small number of deals compared to say desktop licenses.

As per HPC Wire- which is a great tool/newsletter to keep updated on HPC , SAS Institute has been busy on this front partnering with EMC Greenplum and TeraData (who also acquired  SAS Partner AsterData to gain a much needed foot in the MR/SQL space) Continue reading “High Performance Analytics”

R Commander Plugins-20 and growing!

First graphical user interface in 1973.
Image via Wikipedia
R Commander Extensions: Enhancing a Statistical Graphical User Interface by extending menus to statistical packages

R Commander ( see paper by Prof J Fox at http://www.jstatsoft.org/v14/i09/paper ) is a well known and established graphical user interface to the R analytical environment.
While the original GUI was created for a basic statistics course, the enabling of extensions (or plug-ins  http://www.r-project.org/doc/Rnews/Rnews_2007-3.pdf ) has greatly enhanced the possible use and scope of this software. Here we give a list of all known R Commander Plugins and their uses along with brief comments.

  1. DoE – http://cran.r-project.org/web/packages/RcmdrPlugin.DoE/RcmdrPlugin.DoE.pdf
  2. doex
  3. EHESampling
  4. epack- http://cran.r-project.org/web/packages/RcmdrPlugin.epack/RcmdrPlugin.epack.pdf
  5. Export- http://cran.r-project.org/web/packages/RcmdrPlugin.Export/RcmdrPlugin.Export.pdf
  6. FactoMineR
  7. HH
  8. IPSUR
  9. MAc- http://cran.r-project.org/web/packages/RcmdrPlugin.MAc/RcmdrPlugin.MAc.pdf
  10. MAd
  11. orloca
  12. PT
  13. qcc- http://cran.r-project.org/web/packages/RcmdrPlugin.qcc/RcmdrPlugin.qcc.pdf and http://cran.r-project.org/web/packages/qcc/qcc.pdf
  14. qual
  15. SensoMineR
  16. SLC
  17. sos
  18. survival-http://cran.r-project.org/web/packages/RcmdrPlugin.survival/RcmdrPlugin.survival.pdf
  19. SurvivalT
  20. Teaching Demos

Note the naming convention for above e plugins is always with a Prefix of “RCmdrPlugin.” followed by the names above
Also on loading a Plugin, it must be already installed locally to be visible in R Commander’s list of load-plugin, and R Commander loads the e-plugin after restarting.Hence it is advisable to load all R Commander plugins in the beginning of the analysis session.

However the notable E Plugins are
1) DoE for Design of Experiments-
Full factorial designs, orthogonal main effects designs, regular and non-regular 2-level fractional
factorial designs, central composite and Box-Behnken designs, latin hypercube samples, and simple D-optimal designs can currently be generated from the GUI. Extensions to cover further latin hypercube designs as well as more advanced D-optimal designs (with blocking) are planned for the future.
2) Survival- This package provides an R Commander plug-in for the survival package, with dialogs for Cox models, parametric survival regression models, estimation of survival curves, and testing for differences in survival curves, along with data-management facilities and a variety of tests, diagnostics and graphs.
3) qcc -GUI for  Shewhart quality control charts for continuous, attribute and count data. Cusum and EWMA charts. Operating characteristic curves. Process capability analysis. Pareto chart and cause-and-effect chart. Multivariate control charts
4) epack- an Rcmdr “plug-in” based on the time series functions. Depends also on packages like , tseries, abind,MASS,xts,forecast. It covers Log-Exceptions garch
and following Models -Arima, garch, HoltWinters
5)Export- The package helps users to graphically export Rcmdr output to LaTeX or HTML code,
via xtable() or Hmisc::latex(). The plug-in was originally intended to facilitate exporting Rcmdr
output to formats other than ASCII text and to provide R novices with an easy-to-use,
easy-to-access reference on exporting R objects to formats suited for printed output. The
package documentation contains several pointers on creating reports, either by using
conventional word processors or LaTeX/LyX.
6) MAc- This is an R-Commander plug-in for the MAc package (Meta-Analysis with
Correlations). This package enables the user to conduct a meta-analysis in a menu-driven,
graphical user interface environment (e.g., SPSS), while having the full statistical capabilities of
R and the MAc package. The MAc package itself contains a variety of useful functions for
conducting a research synthesis with correlational data. One of the unique features of the MAc
package is in its integration of user-friendly functions to complete the majority of statistical steps
involved in a meta-analysis with correlations. It uses recommended procedures as described in
The Handbook of Research Synthesis and Meta-Analysis (Cooper, Hedges, & Valentine, 2009).

A query to help for ??Rcmdrplugins reveals the following information which can be quite overwhelming given that almost 20 plugins are now available-

RcmdrPlugin.DoE::DoEGlossary
Glossary for DoE terminology as used in
RcmdrPlugin.DoE
RcmdrPlugin.DoE::Menu.linearModelDesign
RcmdrPlugin.DoE Linear Model Dialog for
experimental data
RcmdrPlugin.DoE::Menu.rsm
RcmdrPlugin.DoE response surface model Dialog
for experimental data
RcmdrPlugin.DoE::RcmdrPlugin.DoE-package
R-Commander plugin package that implements
design of experiments facilities from packages
DoE.base, FrF2 and DoE.wrapper into the
R-Commander
RcmdrPlugin.DoE::RcmdrPlugin.DoEUndocumentedFunctions
Functions used in menus
RcmdrPlugin.doex::ranblockAnova
Internal RcmdrPlugin.doex objects
RcmdrPlugin.doex::RcmdrPlugin.doex-package
Install the DOEX Rcmdr Plug-In
RcmdrPlugin.EHESsampling::OpenSampling1
Internal functions for menu system of
RcmdrPlugin.EHESsampling
RcmdrPlugin.EHESsampling::RcmdrPlugin.EHESsampling-package
Help with EHES sampling
RcmdrPlugin.Export::RcmdrPlugin.Export-package
Graphically export objects to LaTeX or HTML
RcmdrPlugin.FactoMineR::defmacro
Internal RcmdrPlugin.FactoMineR objects
RcmdrPlugin.FactoMineR::RcmdrPlugin.FactoMineR
Graphical User Interface for FactoMineR
RcmdrPlugin.IPSUR::IPSUR-package
An IPSUR Plugin for the R Commander
RcmdrPlugin.MAc::RcmdrPlugin.MAc-package
Meta-Analysis with Correlations (MAc) Rcmdr
Plug-in
RcmdrPlugin.MAd::RcmdrPlugin.MAd-package
Meta-Analysis with Mean Differences (MAd) Rcmdr
Plug-in
RcmdrPlugin.orloca::activeDataSetLocaP
RcmdrPlugin.orloca: A GUI for orloca-package
(internal functions)
RcmdrPlugin.orloca::RcmdrPlugin.orloca-package
RcmdrPlugin.orloca: A GUI for orloca-package
RcmdrPlugin.orloca::RcmdrPlugin.orloca.es
RcmdrPlugin.orloca.es: Una interfaz grafica
para el paquete orloca
RcmdrPlugin.qcc::RcmdrPlugin.qcc-package
Install the Demos Rcmdr Plug-In
RcmdrPlugin.qual::xbara
Internal RcmdrPlugin.qual objects
RcmdrPlugin.qual::RcmdrPlugin.qual-package
Install the quality Rcmdr Plug-In
RcmdrPlugin.SensoMineR::defmacro
Internal RcmdrPlugin.SensoMineR objects
RcmdrPlugin.SensoMineR::RcmdrPlugin.SensoMineR
Graphical User Interface for SensoMineR
RcmdrPlugin.SLC::Rcmdr.help.RcmdrPlugin.SLC
RcmdrPlugin.SLC: A GUI for slc-package
(internal functions)
RcmdrPlugin.SLC::RcmdrPlugin.SLC-package
RcmdrPlugin.SLC: A GUI for SLC R package
RcmdrPlugin.sos::RcmdrPlugin.sos-package
Efficiently search R Help pages
RcmdrPlugin.steepness::Rcmdr.help.RcmdrPlugin.steepness
RcmdrPlugin.steepness: A GUI for
steepness-package (internal functions)
RcmdrPlugin.steepness::RcmdrPlugin.steepness
RcmdrPlugin.steepness: A GUI for steepness R
package
RcmdrPlugin.survival::allVarsClusters
Internal RcmdrPlugin.survival Objects
RcmdrPlugin.survival::RcmdrPlugin.survival-package
Rcmdr Plug-In Package for the survival Package
RcmdrPlugin.TeachingDemos::RcmdrPlugin.TeachingDemos-package
Install the Demos Rcmdr Plug-In

 

Professors and Patches: For a Betterrrr R

Professors sometime throw out provocative statements to ensure intellectual debate. I have had almost 1500+ hits in less than 2 days ( and I am glad I am on wordpress.com , my old beloved server would have crashed))

The remarks from Ross Ihaka, covered before and also at Xian’s blog at

Note most of his remarks are techie- and only a single line refers to Revlution Analytics.

Other senior members of community (read- professors are silent, though brobably some thought may have been ignited behind scenes)

http://xianblog.wordpress.com/2010/09/06/insane/comment-page-4/#comments

Ross Ihaka Says:
September 12, 2010 at 1:23 pm

Since (something like) my name has been taken in vain here, let me
chip in.

I’ve been worried for some time that R isn’t going to provide the base
that we’re going to need for statistical computation in the
future. (It may well be that the future is already upon us.) There
are certainly efficiency problems (speed and memory use), but there
are more fundamental issues too. Some of these were inherited from S
and some are peculiar to R.

One of the worst problems is scoping. Consider the following little
gem.

f =
function() {
if (runif(1) > .5)
x = 10
x
}

The x being returned by this function is randomly local or global.
There are other examples where variables alternate between local and
non-local throughout the body of a function. No sensible language
would allow this. It’s ugly and it makes optimisation really
difficult. This isn’t the only problem, even weirder things happen
because of interactions between scoping and lazy evaluation.

In light of this, I’ve come to the conclusion that rather than
“fixing” R, it would be much more productive to simply start over and
build something better. I think the best you could hope for by fixing
the efficiency problems in R would be to boost performance by a small
multiple, or perhaps as much as an order of magnitude. This probably
isn’t enough to justify the effort (Luke Tierney has been working on R
compilation for over a decade now).

To try to get an idea of how much speedup is possible, a number of us
have been carrying out some experiments to see how much better we
could do with something new. Based on prototyping we’ve been doing at
Auckland, it looks like it should be straightforward to get two orders
of magnitude speedup over R, at least for those computations which are
currently bottle-necked. There are a couple of ways to make this
happen.

First, scalar computations in R are very slow. This in part because
the R interpreter is very slow, but also because there are a no scalar
types. By introducing scalars and using compilation it looks like its
possible to get a speedup by a factor of several hundred for scalar
computations. This is important because it means that many ghastly
uses of array operations and the apply functions could be replaced by
simple loops. The cost of these improvements is that scope
declarations become mandatory and (optional) type declarations are
necessary to help the compiler.

As a side-effect of compilation and the use of type-hinting it should
be possible to eliminate dispatch overhead for certain (sealed)
classes (scalars and arrays in particular). This won’t bring huge
benefits across the board, but it will mean that you won’t have to do
foreign language calls to get efficiency.

A second big problem is that computations on aggregates (data frames
in particular) run at glacial rates. This is entirely down to
unnecessary copying because of the call-by-value semantics.
Preserving call-by-value semantics while eliminating the extra copying
is hard. The best we can probably do is to take a conservative
approach. R already tries to avoid copying where it can, but fails in
an epic fashion. The alternative is to abandon call-by-value and move
to reference semantics. Again, prototyping indicates that several
hundredfold speedup is possible (for data frames in particular).

The changes in semantics mentioned above mean that the new language
will not be R. However, it won’t be all that far from R and it should
be easy to port R code to the new system, perhaps using some form of
automatic translation.

If we’re smart about building the new system, it should be possible to
make use of multi-cores and parallelism. Adding this to the mix might just
make it possible to get a three order-of-magnitude performance boost
with just a fraction of the memory that R uses. I think it’s something
really worth putting some effort into.

I also think one other change is necessary. The license will need to a
better job of protecting work donated to the commons than GPL2 seems
to have done. I’m not willing to have any more of my work purloined by
the likes of Revolution Analytics, so I’ll be looking for better
protection from the license (and being a lot more careful about who I
work with).

The discussion spilled over to Stack Overflow as well

http://stackoverflow.com/questions/3706990/is-r-that-bad-that-it-should-be-rewritten-from-scratch/3710667#3710667

n the past week I’ve been following a discussion where Ross Ihaka wrote (here ):

I’ve been worried for some time that R isn’t going to provide the base that we’re going to need for statistical computation in the future. (It may well be that the future is already upon us.) There are certainly efficiency problems (speed and memory use), but there are more fundamental issues too. Some of these were inherited from S and some are peculiar to R.

He then continued explaining. This discussion started from this post, and was then followed by commentsherehereherehereherehere and maybe some more places I don’t know of.

We all know the problem now.

R can be improved substantially in terms of speed.

For some solutions, here are the patches by Radford-

http://www.cs.toronto.edu/~radford/speed-patches-doc

patch-dollar

    Speeds up access to lists, pairlists, and environments using the
    $ operator.  The speedup comes mainly from avoiding the overhead of 
    calling DispatchOrEval if there are no complexities, from passing
    on the field to extract as a symbol, or a name, or both, as available,
    and then converting only as necessary, from simplifying and inlining
    the pstrmatch procedure, and from not translating string multiple
    times.  

    Relevant timing test script:  test-dollar.r 

    This test shows about a 40% decrease in the time needed to extract
    elements of lists and environments.

    Changes unrelated to speed improvement:

    A small error-reporting bug is fixed, illustrated by the following
    output with r52822:

    > options(warnPartialMatchDollar=TRUE)
    > pl <- pairlist(abc=1,def=2)
    > pl$ab
    [1] 1
    Warning message:
    In pl$ab : partial match of 'ab' to ''

    Some code is changed at the end of R_subset3_dflt because it seems 
    to be more correct, as discussed in code comments. 

patch-evalList

    Speeds up a large number of operations by avoiding allocation of
    an extra CONS cell in the procedures for evaluating argument lists.

    Relevant timing test scripts:  all of them, but will look at test-em.r 

    On test-em.r, the speedup from this patch is about 5%.

patch-fast-base

    Speeds up lookup of symbols defined in the base environment, by
    flagging symbols that have a base environment definition recorded
    in the global cache.  This allows the definition to be retrieved
    quickly without looking in the hash table.  

    Relevant timing test scripts:  all of them, but will look at test-em.r 

    On test-em.r, the speedup from this patch is about 3%.

    Issue:  This patch uses the "spare" bit for the flag.  This bit is
    misnamed, since it is already used elsewhere (for closures).  It is
    possible that one of the "gp" bits should be used instead.  The
    "gp" bits should really be divided up for faster access, and so that
    their present use is apparent in the code.

    In case this use of the "spare" bit proves unwise, the patch code is 
    conditional on FAST_BASE_CACHE_LOOKUP being defined at the start of
    envir.r.

patch-fast-spec

    Speeds up lookup of function symbols that begin with a character
    other than a letter or ".", by allowing fast bypass of non-global
    environments that do not contain (and have never contained) symbols 
    of this sort.  Since it is expected that only functions will be
    given names of this sort, the check is done only in findFun, though
    it could also be done in findVar.

    Relevant timing test scripts:  all of them, but will look at test-em.r 

    On test-em.r, the speedup from this patch is about 8%.    

    Issue:  This patch uses the "spare" bit to flag environments known
    to not have symbols starting with a special character.  See remarks
    on patch-fast-base.

    In case this use of the "spare" bit proves unwise, the patch code is 
    conditional on FAST_SPEC_BYPASS being defined at the start of envir.r.

patch-for

    Speeds up for loops by not allocating new space for the loop
    variable every iteration, unless necessary.  

    Relevant timing test script:  test-for.r

    This test shows a speedup of about 5%.  

    Change unrelated to speed improvement:

    Fixes what I consider to be a bug, in which the loop clobbers a
    global variable, as demonstrated by the following output with r52822:

    > i <- 99
    > f <- function () for (i in 1:3) { print(i); if (i==2) rm(i); }
    > f()
    [1] 1
    [1] 2
    [1] 3
    > print(i)
    [1] 3

patch-matprod

    Speeds up matrix products, including vector dot products.  The
    speed issue here is that the R code checks for any NAs, and 
    does the multiply in the matprod procedure (in array.c) if so,
    since BLAS isn't trusted with NAs.  If this check takes about
    as long as just doing the multiply in matprod, calling a BLAS
    routine makes no sense.  

    Relevant time test script:  test-matprod.r

    With no external BLAS, this patch speeds up long vector-vector 
    products by a factor of about six, matrix-vector products by a
    factor of about three, and some matrix-matrix products by a 
    factor of about two.

    Issue:  The matrix multiply code in matprod using an LDOUBLE
    (long double) variable to accumulate sums, for improved accuracy.  
    On a SPARC system I tested on, operations on long doubles are 
    vastly slower than on doubles, so that the patch produces a 
    large slowdown rather than an improvement.  This is also an issue 
    for the "sum" function, which also uses an LDOUBLE to accumulate
    the sum.  Perhaps an ordinarly double should be used in these
    places, or perhaps the configuration script should define LDOUBLE 
    as double on architectures where long doubles are extraordinarily 
    slow.

    Due to this issue, not defining MATPROD_CAN_BE_DONE_HERE at the
    start of array.c will disable this patch.

patch-parens

    Speeds up parentheses by making "(" a special operator whose
    argument is not evaluated, thereby bypassing the overhead of
    evalList.  Also slightly speeds up curly brackets by inlining
    a function that is stylistically better inline anyway.

    Relevant test script:  test-parens.r

    In the parens part of test-parens.r, the speedup is about 9%.

patch-protect

    Speeds up numerous operations by making PROTECT, UNPROTECT, etc.
    be mostly macros in the files in src/main.  This takes effect
    only for files that include Defn.h after defining the symbol
    USE_FAST_PROTECT_MACROS.  With these macros, code of the form
    v = PROTECT(...) must be replaced by PROTECT(v = ...).  

    Relevant timing test scripts:  all of them, but will look at test-em.r 

    On test-em.r, the speedup from this patch is about 9%.

patch-save-alloc

    Speeds up some binary and unary arithmetic operations by, when
    possible, using the space holding one of the operands to hold
    the result, rather than allocating new space.  Though primarily
    a speed improvement, for very long vectors avoiding this allocation 
    could avoid running out of space.

    Relevant test script:  test-complex-expr.r

    On this test, the speedup is about 5% for scalar operands and about
    8% for vector operands.

    Issues:  There are some tricky issues with attributes, but I think
    I got them right.  This patch relies on NAMED being set correctly 
    in the rest of the code.  In case it isn't, the patch can be disabled 
    by not defining AVOID_ALLOC_IF_POSSIBLE at the top of arithmetic.c.

patch-square

    Speeds up a^2 when a is a long vector by not checking for the
    special case of an exponent of 2 over and over again for every 
    vector element.

    Relevant test script:  test-square.r

    The time for squaring a long vector is reduced in this test by a
    factor of more than five.

patch-sum-prod

    Speeds up the "sum" and "prod" functions by not checking for NA
    when na.rm=FALSE, and other detailed code improvements.

    Relevant test script:  test-sum-prod.r

    For sum, the improvement is about a factor of 2.5 when na.rm=FALSE,
    and about 10% when na.rm=TRUE.

    Issue:  See the discussion of patch-matprod regarding LDOUBLE.
    There is no change regarding this issue due to this patch, however.

patch-transpose

    Speeds up the transpose operation (the "t" function) from detailed
    code improvements.

    Relevant test script:  test-transpose.r

    The improvement for 200x60 matrices is about a factor of two.
    There is little or no improvement for long row or column vectors.

patch-vec-arith

    Speeds up arithmetic on vectors of the same length, or when on
    vector is of length one.  This is done with detailed code improvements.

    Relevant test script:  test-vec-arith.r

    On long vectors, the +, -, and * operators are sped up by about     
    20% when operands are the same length or one operand is of length one.

    Rather mysteriously, when the operands are not length one or the
    same length, there is about a 20% increase in time required, though
    this may be due to some strange C optimizer peculiarity or some 
    strange cache effect, since the C code for this is the same as before,
    with negligible additional overhead getting to it.  Regardless, this 
    case is much less common than equal lengths or length one.

    There is little change for the / operator, which is much slower than
    +, -, or *.

patch-vec-subset

    Speeds up extraction of subsets of vectors or matrices (eg, v[10:20]
    or M[1:10,101:110]).  This is done with detailed code improvements.

    Relevant test script:  test-vec-subset.r

    There are lots of tests in this script.  The most dramatic improvement
    is for extracting many rows and columns of a large array, where the 
    improvement is by about a factor of four.  Extracting many rows from
    one column of a matrix is sped up by about 30%. 

    Changes unrelated to speed improvement:

    Fixes two latent bugs where the code incorrectly refers to NA_LOGICAL
    when NA_INTEGER is appropriate and where LOGICAL and INTEGER types
    are treated as interchangeable.  These cause no problems at the moment,
    but would if representations were changed.

patch-subscript

    (Formerly part of patch-vec-subset)  This patch also speeds up
    extraction, and also replacement, of subsets of vectors or
    matrices, but focuses on the creation of the indexes rather than
    the copy operations.  Often avoids a duplication (see below) and
    eliminates a second scan of the subscript vector for zero
    subscripts, folding it into a previous scan at no additional cost.

    Relevant test script:  test-vec-subset.r

    Speeds up some operations with scalar or short vector indexes by
    about 10%.  Speeds up subscripting with a longer vector of
    positive indexes by about 20%.

    Issues:  The current code duplicates a vector of indexes when it
    seems unnecessary.  Duplication is for two reasons:  to handle
    the situation where the index vector is itself being modified in
    a replace operation, and so that any attributes can be removed, which 
    is helpful only for string subscripts, given how the routine to handle 
    them returns information via an attribute.  Duplication for the
    second reasons can easily be avoided, so I avoided it.  The first
    reason for duplication is sometimes valid, but can usually be avoided
    by first only doing it if the subscript is to be used for replacement
    rather than extraction, and second only doing it if the NAMED field
    for the subscript isn't zero.

    I also removed two layers of procedure call overhead (passing seven
    arguments, so not trivial) that seemed to be doing nothing.  Probably 
    it used to do something, but no longer does, but if instead it is 
    preparation for some future use, then removing it might be a mistake.

Software problems are best solved by writing code or patches in my opinion rather than discussing endlessly
Some other solutions to a BETTERRRR R
1) Complete Code Design Review
2) Version 3 - Tuneup
3) Better Documentation
4) Suing Revolution Analytics for the code - Hand over da code pardner

Towards better analytical software

Here are some thoughts on using existing statistical software for better analytics and/or business intelligence (reporting)-

1) User Interface Design Matters- Most stats software have a legacy approach to user interface design. While the Graphical User Interfaces need to more business friendly and user friendly- example you can call a button T Test or You can call it Compare > Means of Samples (with a highlight called T Test). You can call a button Chi Square Test or Call it Compare> Counts Data. Also excessive reliance on drop down ignores the next generation advances in OS- namely touchscreen instead of mouse click and point.

Given the fact that base statistical procedures are the same across softwares, a more thoughtfully designed user interface (or revamped interface) can give softwares an edge over legacy designs.

2) Branding of Software Matters- One notable whine against SAS Institite products is a premier price. But really that software is actually inexpensive if you see other reporting software. What separates a Cognos from a Crystal Reports to a SAS BI is often branding (and user interface design). This plays a role in branding events – social media is often the least expensive branding and marketing channel. Same for WPS and Revolution Analytics.

3) Alliances matter- The alliances of parent companies are reflected in the sales of bundled software. For a complete solution , you need a database plus reporting plus analytical software. If you are not making all three of the above, you need to partner and cross sell. Technically this means that software (either DB, or Reporting or Analytics) needs to talk to as many different kinds of other softwares and formats. This is why ODBC in R is important, and alliances for small companies like Revolution Analytics, WPS and Netezza are just as important as bigger companies like IBM SPSS, SAS Institute or SAP. Also tie-ins with Hadoop (like R and Netezza appliance)  or  Teradata and SAS help create better usage.

4) Cloud Computing Interfaces could be the edge- Maybe cloud computing is all hot air. Prudent business planing demands that any software maker in analytics or business intelligence have an extremely easy to load interface ( whether it is a dedicated on demand website) or an Amazon EC2 image. Easier interfaces win and with the cloud still in early stages can help create an early lead. For R software makers this is critical since R is bad in PC usage for larger sets of data in comparison to counterparts. On the cloud that disadvantage vanishes. An easy to understand cloud interface framework is here ( its 2 years old but still should be okay) http://knol.google.com/k/data-mining-through-cloud-computing#

5) Platforms matter- Softwares should either natively embrace all possible platforms or bundle in middle ware themselves.

Here is a case study SAS stopped supporting Apple OS after Base SAS 7. Today Apple OS is strong  ( 3.47 million Macs during the most recent quarter ) and the only way to use SAS on a Mac is to do either

http://goo.gl/QAs2

or do a install of Ubuntu on the Mac ( https://help.ubuntu.com/community/MacBook ) and do this

http://ubuntuforums.org/showthread.php?t=1494027

Why does this matter? Well SAS is free to academics and students  from this year, but Mac is a preferred computer there. Well WPS can be run straight away on the Mac (though they are curiously not been able to provide academics or discounted student copies 😉 ) as per

http://goo.gl/aVKu

Does this give a disadvantage based on platform. Yes. However JMP continues to be supported on Mac. This is also noteworthy given the upcoming Chromium OS by Google, Windows Azure platform for cloud computing.