Here is an interview with Charlie Parker, head of large scale online algorithms at http://bigml.com
Ajay- Describe your own personal background in scientific computing, and how you came to be involved with machine learning, cloud computing and BigML.com
Charlie- I am a machine learning Ph.D. from Oregon State University. Francisco Martin (our founder and CEO), Adam Ashenfelter (the lead developer on the tree algorithm), and myself were all studying machine learning at OSU around the same time. We all went our separate ways after that.
Francisco started Strands and turned it into a 100+ million dollar company building recommender systems. Adam worked for CleverSet, a probabilistic modeling company that was eventually sold to Cisco, I believe. I worked for several years in the research labs at Eastman Kodak on data mining, text analysis, and computer vision.
When Francisco left Strands to start BigML, he brought in Justin Donaldson who is a brilliant visualization guy from Indiana, and an ex-Googler named Jose Ortega who is responsible for most of our data infrastructure. They pulled in Adam and I a few months later. We also have Poul Petersen, a former Strands employee, who manages our herd of servers. He is a wizard and makes everyone else’s life much easier.
Ajay- You use clojure for the back end of BigML.com .Are there any other languages and packages you are considering? What makes clojure such a good fit for cloud computing ?
Charlie- Clojure is a great language because it offers you all of the benefits of Java (extensive libraries, cross-platform compatibility, easy integration with things like Hadoop, etc.) but has the syntactical elegance of a functional language. This makes our code base small and easy to read as well as powerful.
We’ve had occasional issues with speed, but that just means writing the occasional function or library in Java. As we build towards processing data at the Terabyte level, we’re hoping to create a framework that is language-agnostic to some extent. So if we have some great machine learning code in C, for example, we’ll use Clojure to tie everything together, but the code that does the heavy lifting will still be in C. For the API and Web layers, we use Python and Django, and Justin is a huge fan of HaXe for our visualizations.
Ajay- Current support is for Decision Trees. When can we see SVM, K Means Clustering and Logit Regression?
Charlie- Right now we’re focused on perfecting our infrastructure and giving you new ways to put data in the system, but expect to see more algorithms appearing in the next few months. We want to make sure they are as beautiful and easy to use as the trees are. Without giving too much away, the first new thing we will probably introduce is an ensemble method of some sort (such as Boosting or Bagging). Clustering is a little further away but we’ll get there soon!
Ajay- How can we use the BigML.com API using R and Python.
Charlie- We have a public github repo for the language bindings. https://github.com/bigmlcom/io Right now, there there are only bash scripts but that should change very soon. The python bindings should be there in a matter of days, and the R bindings in probably a week or two. Clojure and Java bindings should follow shortly after that. We’ll have a blog post about it each time we release a new language binding. http://blog.bigml.com/
Ajay- How can we predict large numbers of observations using a Model that has been built and pruned (model scoring)?
Charlie- We are in the process of refactoring our backend right now for better support for batch prediction and model evaluation. This is something that is probably only a few weeks away. Keep your eye on our blog for updates!
Ajay- How can we export models built in BigML.com for scoring data locally.
Charlie- This is as simple as a call to our API. https://bigml.com/developers/models The call gives you a JSON object representing the tree that is roughly equivalent to a PMML-style representation.
Related Articles (Ps the Related Articles is auto generated by Zementa- a software embedded within WordPress.com in case you are wondering what the deal with the linking is)
Just working with PySpread- and worked on a 1 million by 1 million spreadsheet- Python sure looks promising for the way ahead for stat computing ( you need to
Pyspread is a cross-platform Python spreadsheet application. It is based on and written in the programming language Python.
Instead of spreadsheet formulas, Python expressions are entered into the spreadsheet cells. Each expression returns a Python object that can be accessed from other cells. These objects can represent anything including lists or matrices.
features
Three dimensional grid with up to 85,899,345 rows and 14,316,555 columns (64 bit systems, depends on row height and column width). Note that a million cells require about 500 MB of memory.
Complex data types such as lists, trees or matrices within a single cell.
Macros for functionalities that are too complex for a single Python expression.
Python module access from each cell, which allows:
Fixed point decimal numbers for business calculations, (via the decimal module from the standard library)
Advanced statistics including plotting functions (via RPy)
Much more via <your favourite module>.
CSV import and export
Clipboard access
warning
The concept of pyspread allows doing everything from each cell that a Python script can do. This powerful feature has its drawbacks. A spreadsheet may very well delete your hard drive or send your data via the Internet. Of course this is a non-issue if you sandbox properly or if you only use self developed spreadsheets.
Since this is not the case for everyone (see discussion at lwn.net), a GPG signature based trust model for spreadsheet files has been introduced. It ensures that only your own trusted files are executed on loading. Untrusted files are displayed in safe mode. You can approve a file manually. Inspect carefully.
Here is a short list of resources and material I put together as starting points for R and Cloud Computing It’s a bit messy but overall should serve quite comprehensively.
Cloud computing is a commonly used expression to imply a generational change in computing from desktop-servers to remote and massive computing connections,shared computers, enabled by high bandwidth across the internet.
As per the National Institute of Standards and Technology Definition,
Cloud computing is a model for enabling convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications, and services) that can be rapidly provisioned and released with minimal management effort or service provider interaction.
Rweb is developed and maintained by Jeff Banfield. The Rweb Home Page provides access to all three versions of Rweb—a simple text entry form that returns output and graphs, a more sophisticated JavaScript version that provides a multiple window environment, and a set of point and click modules that are useful for introductory statistics courses and require no knowledge of the R language. All of the Rweb versions can analyze Web accessible datasets if a URL is provided.
The paper “Rweb: Web-based Statistical Analysis”, providing a detailed explanation of the different versions of Rweb and an overview of how Rweb works, was published in the Journal of Statistical Software (http://www.jstatsoft.org/v04/i01/).
Ulf Bartel has developed R-Online, a simple on-line programming environment for R which intends to make the first steps in statistical programming with R (especially with time series) as easy as possible. There is no need for a local installation since the only requirement for the user is a JavaScript capable browser. See http://osvisions.com/r-online/ for more information.
Rcgi is a CGI WWW interface to R by MJ Ray. It had the ability to use “embedded code”: you could mix user input and code, allowing the HTMLauthor to do anything from load in data sets to enter most of the commands for users without writing CGI scripts. Graphical output was possible in PostScript or GIF formats and the executed code was presented to the user for revision. However, it is not clear if the project is still active.
Currently, a modified version of Rcgi by Mai Zhou (actually, two versions: one with (bitmap) graphics and one without) as well as the original code are available from http://www.ms.uky.edu/~statweb/.
David Firth has written CGIwithR, an R add-on package available from CRAN. It provides some simple extensions to R to facilitate running R scripts through the CGI interface to a web server, and allows submission of data using both GET and POST methods. It is easily installed using Apache under Linux and in principle should run on any platform that supports R and a web server provided that the installer has the necessary security permissions. David’s paper “CGIwithR: Facilities for Processing Web Forms Using R” was published in the Journal of Statistical Software (http://www.jstatsoft.org/v08/i10/). The package is now maintained by Duncan Temple Lang and has a web page athttp://www.omegahat.org/CGIwithR/.
Rpad, developed and actively maintained by Tom Short, provides a sophisticated environment which combines some of the features of the previous approaches with quite a bit of JavaScript, allowing for a GUI-like behavior (with sortable tables, clickable graphics, editable output), etc.
Jeff Horner is working on the R/Apache Integration Project which embeds the R interpreter inside Apache 2 (and beyond). A tutorial and presentation are available from the project web page at http://biostat.mc.vanderbilt.edu/twiki/bin/view/Main/RApacheProject.
Rserve is a project actively developed by Simon Urbanek. It implements a TCP/IP server which allows other programs to use facilities of R. Clients are available from the web site for Java and C++ (and could be written for other languages that support TCP/IP sockets).
OpenStatServer is being developed by a team lead by Greg Warnes; it aims “to provide clean access to computational modules defined in a variety of computational environments (R, SAS, Matlab, etc) via a single well-defined client interface” and to turn computational services into web services.
Two projects use PHP to provide a web interface to R. R_PHP_Online by Steve Chen (though it is unclear if this project is still active) is somewhat similar to the above Rcgi and Rweb. R-php is actively developed by Alfredo Pontillo and Angelo Mineo and provides both a web interface to R and a set of pre-specified analyses that need no R code input.
webbioc is “an integrated web interface for doing microarray analysis using several of the Bioconductor packages” and is designed to be installed at local sites as a shared computing resource.
Rwui is a web application to create user-friendly web interfaces for R scripts. All code for the web interface is created automatically. There is no need for the user to do any extra scripting or learn any new scripting techniques. Rwui can also be found at http://rwui.cryst.bbk.ac.uk.
Finally, the R.rsp package by Henrik Bengtsson introduces “R Server Pages”. Analogous to Java Server Pages, an R server page is typically HTMLwith embedded R code that gets evaluated when the page is requested. The package includes an internal cross-platform HTTP server implemented in Tcl, so provides a good framework for including web-based user interfaces in packages. The approach is similar to the use of the brew package withRapache with the advantage of cross-platform support and easy installation.
Remote access to R/Bioconductor on EBI’s 64-bit Linux Cluster
Start the workbench by downloading the package for your operating system (Macintosh or Windows), or via Java Web Start, and you will get access to an instance of R running on one of EBI’s powerful machines. You can install additional packages, upload your own data, work with graphics and collaborate with colleagues, all as if you are running R locally, but unlimited by your machine’s memory, processor or data storage capacity.
Most up-to-date R version built for multicore CPUs
Access to all Bioconductor packages
Access to our computing infrastructure
Fast access to data stored in EBI’s repositories (e.g., public microarray data in ArrayExpress)
Using R Google Docs http://www.omegahat.org/RGoogleDocs/run.pdf
It uses the XML and RCurl packages and illustrates that it is relatively quick and easy
to use their primitives to interact with Web services.
Amazon’s EC2 is a type of cloud that provides on demand computing infrastructures called an Amazon Machine Images or AMIs. In general, these types of cloud provide several benefits:
Simple and convenient to use. An AMI contains your applications, libraries, data and all associated configuration settings. You simply access it. You don’t need to configure it. This applies not only to applications like R, but also can include any third-party data that you require.
On-demand availability. AMIs are available over the Internet whenever you need them. You can configure the AMIs yourself without involving the service provider. You don’t need to order any hardware and set it up.
Elastic access. With elastic access, you can rapidly provision and access the additional resources you need. Again, no human intervention from the service provider is required. This type of elastic capacity can be used to handle surge requirements when you might need many machines for a short time in order to complete a computation.
Pay per use. The cost of 1 AMI for 100 hours and 100 AMI for 1 hour is the same. With pay per use pricing, which is sometimes called utility pricing, you simply pay for the resources that you use.
#This example requires you had previously created a bucket named data_language on your Google Storage and you had uploaded a CSV file named language_id.txt (your data) into this bucket – see for details
library(predictionapirwrapper)
Elastic-R is a new portal built using the Biocep-R platform. It enables statisticians, computational scientists, financial analysts, educators and students to use cloud resources seamlessly; to work with R engines and use their full capabilities from within simple browsers; to collaborate, share and reuse functions, algorithms, user interfaces, R sessions, servers; and to perform elastic distributed computing with any number of virtual machines to solve computationally intensive problems.
Also see Karim Chine’s http://biocep-distrib.r-forge.r-project.org/
R for Salesforce.com
At the point of writing this, there seem to be zero R based apps on Salesforce.com This could be a big opportunity for developers as both Apex and R have similar structures Developers could write free code in R and charge for their translated version in Apex on Salesforce.com
Force.com and Salesforce have many (1009) apps at http://sites.force.com/appexchange/home for cloud computing for
businesses, but very few forecasting and statistical simulation apps.
These are like iPhone apps except meant for business purposes (I am
unaware if any university is offering salesforce.com integration
though google apps and amazon related research seems to be on)
Personal Note-Mentioning SAS in an email to a R list is a big no-no in terms of getting a response and love. Same for being careless about which R help list to email (like R devel or R packages or R help)
Pyspread is a cross-platform Python spreadsheet application. It is based on and written in the programming language Python.
Instead of spreadsheet formulas, Python expressions are entered into the spreadsheet cells. Each expression returns a Python object that can be accessed from other cells. These objects can represent anything including lists or matrices.
features
In pyspread, cells expect Python expressions and return Python objects. Therefore, complex data types such as lists, trees or matrices can be handled within a single cell. Macros can be used for functions that are too complex for a single expression.
Since Python modules can be easily used without external scripts, arbitrary size rational numbers (via gmpy), fixed point decimal numbers for business calculations, (via the decimal module from the standard library) and advanced statistics including plotting functions (via RPy) can be used in the spreadsheet. Everything is directly available from each cell. Just use the grid
Data can be imported and exported using csv files or the clipboard. Other forms of data exchange is possible using external Python modules.
In order to simplify sparse matrix editing, pyspread features a three dimensional grid that can be sized up to 85,899,345 rows and 14,316,555 columns (64 bit-systems, depends on row height and column width). Note that importing a million cells requires about 500 MB of memory.
The concept of pyspread allows doing everything from each cell that a Python script can do. This may very well include deleting your hard drive or sending your data via the Internet. Of course this is a non-issue if you sandbox properly or if you only use self developed spreadsheets. Since this is not the case for everyone (see the discussion at lwn.net), a GPG signature based trust model for spreadsheet files has been introduced. It ensures that only your own trusted files are executed on loading. Untrusted files are displayed in safe mode. You can trust a file manually. Inspect carefully.
requirements
Pyspread runs on Linux, Windows and *nix platforms with GTK+ support. There are reports that it works with MacOS X as well. If you would like to contribute by testing on OS X please contact me.
a spreadsheet with more powerful functions and data structures that are accessible inside each cell. Something like Python that empowers you to do things quickly. And yes, it should be free and it should run on Linux as well as on Windows. I looked around and found nothing that suited me. Therefore, I started pyspread.
Concept
Each cell accepts any input that works in a Python command line.
The inputs are parsed and evaluated by Python’s eval command.
The result objects are accessible via a 3D numpy object array.
String representations of the result objects are displayed in the cells.
Benefits
Each cell returns a Python object. This object can be anything including arrays and third party library objects.
Generator expressions can be used efficiently for data manipulation.
Efficient numpy slicing is used.
numpy methods are accessible for the data.
Installation
Download the pyspread tarball or zip and unzip at a convenient place
In case you do not have it already get and install Python, wxpython and numpy
If you want the examples to work, install gmpy, R and rpy
I am currently playing/ trying out RApache- one more excellent R product from Vanderbilt’s excellent Dept of Biostatistics and it’s prodigious coder Jeff Horner.
I really liked the virtual machine idea- you can download a virtual image of Rapache and play with it- .vmx is easy to create and great to share-
Basically using R Apache (with an EC2 on backend) can help you create customized dashboards, BI apps, etc all using R’s graphical and statistical capabilities.
Rapache embeds the R interpreter inside the Apache 2 web server. By doing this, Rapache realizes the full potential of R and its facilities over the web. R programmers configure appache by mapping Universal Resource Locaters (URL’s) to either R scripts or R functions. The R code relies on CGI variables to read a client request and R’s input/output facilities to write the response.
One advantage to Rapache’s architecture is robust multi-process management by Apache. In contrast to Rserve and RSOAP, Rapache is a pre-fork server utilizing HTTP as the communications protocol. Another advantage is a clear separation, a loose coupling, of R code from client code. With Rserve and RSOAP, the client must send data and R commands to be executed on the server. With Rapache the only client requirements are the ability to communicate via HTTP. Additionally, Rapache gains significant authentication, authorization, and encryption mechanism by virtue of being embedded in Apache.
Existing Demos of Architechture based on R Apache-
You can download version 1.1.10 of rApache now. There
are only two significant changes and you don’t have to edit your
apache config or change any code (just recompile rApache and
reinstall):
1) Error reporting should be more informative. both when you
accidentally introduce errors in the Apache config, and when your code
introduces warnings and errors from web requests.
I’ve struggled with this one for awhile, not really knowing what
strategy would be best. Basically, rApache hooks into the R I/O layer
at such a low level that it’s hard to capture all warnings and errors
as they occur and introduce them to the user in a sane manner. In
prior releases, when ROutputErrors was in effect (either the apache
directive or the R function) one would typically see a bunch of grey
boxes with a red outline with a title of RApache Warning/Error!!!.
Unfortunately those grey boxes could contain empty lines, one line of
error, or a few that relate to the lines in previously displayed
boxes. Really a big uninformative mess.
The new approach is to print just one warning box with the title
“”Oops!!! <b>rApache</b> has something to tell you. View source and
read the HTML comments at the end.” and then as the title implies you
can read the HTML comment located at the end of the file… after the
closing html. That way, you’re actually reading how R would present
the warnings and errors to you as if you executed the code at the R
command prompt. And if you don’t use ROutputErrors, the warning/error
messages are printed in the Apache log file, just as they were before,
but nicer 😉
2) Code dispatching has changed so please let me know if I’ve
introduced any strange behavior.
This was necessary to enhance error reporting. Prior to this release,
rApache would use R’s C API exclusively to build up the call to your
code that is then passed to R’s evaluation engine. The advantage to
this approach is that it’s much more efficient as there is no parsing
involved, however all information about parse errors, files which
produced errors, etc. were lost. The new approach uses R’s built-in
parse function to build up the call and then passes it of to R. A
slight overhead, but it should be negligible. So, if you feel that
this approach is too slow OR I’ve introduced bugs or strange behavior,
please let me know.
FUTURE PLANS
I’m gaining more experience building Debian/Ubuntu packages each day,
so hopefully by some time in 2011 you can rely on binary releases for
these distributions and not install rApache from source! Fingers
crossed!
Development on the rApache 1.1 branch will be winding down (save bug
fix releases) as I transition to the 1.2 branch. This will involve
taking out a small chunk of code that defines the rApache development
environment (all the CGI variables and the functions such as
setHeader, setCookie, etc) and placing it in its own R package…
unnamed as of yet. This is to facilitate my development of the ralite
R package, a small single user cross-platform web server.
The goal for ralite is to speed up development of R web applications,
take out a bit of friction in the development process by not having to
run the full rApache server. Plus it would allow users to develop in
the rApache enronment while on windows and later deploy on more
capable server environments. The secondary goal for ralite is it’s use
in other web server environments (nginx and IIS come to mind) as a
persistent per-client process.
And finally, wiki.rapache.net will be the new www.rapache.net once I
translate the manual over… any day now.
DECISION STATS 
