Statistical Software Resources on the
Collected by Jim Linnemann
Michigan State University Physics
Completeness or Authoritativeness isn't even a goal--just useful
If I'm missing a good link (your collection, for example), or a link died, please
email it to me!
A word on selection
I sampled links on most top-level pages, and included pages I
thought practicing High Energy Physicsits and Astrophysicists would find useful;
that didn't trap me on their web page; and had a reasonable proportion of live
links. The web is wonderful, but ephemeral; I'm sure you'll find links that
weren't broken when I tried. The opinions are my own; a bit on statistics also
slipped in among the software.
High Energy Physics
- The phystat.org site
is a repository for HEP statistical software, with pointers to the
phystat conference series; see StatPatternRecognition there for
well-tuned multivariate algorithms.
Data Group Statistics Summary describes statistical methods (theory) on
which there is consensus in HEP
- Glen Cowan's
statistical resources page (Royal Holloway physics); go up a link for some
software associated with his book.
- There are some statistical
routines in Root (an interactive data
analysis framerwork). See also roostats and tmva for more useful software in the Root framework. Older libraries include cernlib,
points to other HEP analysis software (including JAS, Java Analysis Studio),
but does not have a specific statistics section
statistics committee a Tevatron experiment's statistics page: mostly methods
simple version of the D0 experiment's Bayesian limit calculator
statistics working group a SLAC experiment's statistics page: methods
and a few applets
- Geant statistical
packages, Maria Grazia Pia, HEP, INFN Genova, C++ library
- Fermilab Advanced Analysis
- gnu gsl (gnu scientific library)
contains random number generators, as well as some histogramming, ntuples,
moments for weighted events, and autocorrelation calculations.
- sourceforge.net a broad repository
of open source software. Basic browsing or search by name without subscribing.
You could troll about in the scientific/engineering section and find, for
- The Computer Physics Communications program
library contains a few items of interest; it requires a subscription to
- A glossary
to help translate from statistics-speak to physics-speak (from one of the phystat conferences).
- Statcodes Eric Feigelson et.
al., Penn State: big collection, with commentary; see also his Astrostatistics
book. Look here--much broader than astsrophysics! Includes
link to web-based VOSTAT (Virtual Observatory Statistics) project, largely
implemented in R (see below).
- StatPy: Python
interfaces to statistical software, Tom Loredo, Cornell; see also his
Inference in the Physical Sciences (Software Section) see especially the
(heavily used by statisticians), and BAYESPACK
Barry Madore, Cal Tech
- Mutual translation glossaries
for astronomers and statisticians, and software, and other goodies from statistician David van Dyk.
Carnegie Mellon's StatLib: a key resource
- Free software and interactive pages from John Pezzullo (retired, Georgetown
- Statistics on the Web
from Clay Hellberg of SPSS
Use your browser to search for Resources to get to the good stuff
- Journal of Statistical Software;
in many programming languages.
bugs Markov Chain MC package
- http://stat.duke.edu/comp/software/ Duke Statistics (formatting is dodgy on page, alas)
- From national labs:
- http://gams.nist.gov/ see Class L for
a mixture of commercial and academic software
e-Handbook of Statistical Methods (Engineering Statistics reference, but
not much on multidimensional data, and little software under Tools and Aids)
- There is a wiki list of statistical software
- And finally, a handy statistics glossary.
Statistical Computations in Java on
Some nice things, some trivia, and many broken links. Gives a feel for
the strengths and limitations of web-interfaced statistics. Pezzulo's page above is the best starting place. Many java
links are powered by sisa
and Statistical Learning
Useful buzzwords to search on in bold; "statistics"
will get you more data than methods. Try wiki as well as search engines.
- R The R project for Statistical
Computing: gnu implementation of the S language
- Graphics, statistical algorithms, and a huge repository (CRAN)
of R packages. Extensive online documentation. Published books include Introductory
Statistics with R by Dalgaard; and Programming with Data: A Guide to the S
Language by Chambers; Modern Applied Statistics with S-PLUS, by Venables &
Ripley, and others; here's a very good R
tutorial ; for more, search for "tutorial using R"
- http://www.ggobi.org/ GGobi visualization
package for multidimensional data.
- Includes dynamic graphics such as arrays of scatterplots, brushing techniques
(highlighting groups of objectes in one dimension and having their coordinates
highlighted in other coordinates); parallel coordinate plots, and grand tours.
Interfaces exist to R and Python front ends, and database back ends. I've
skimped on Perl here and elsewhere, but often where you find Python
interfaces, you'll also find Perl-and sometimes Ruby.
- http://www.omegahat.org/ The Omega
project for Statistical Computing.
- Interfaces between R, Python, XML, Java, databases, and other goodies.
At this point, aimed more at developers than users.
- Jerry Friedman (High
Energy Physicist turned Statistican) has software for a number of multivariate
techniques on the web; don't miss his book below.
Elements of Statistical Learning Theory, by Hastie, Tibshirani, and Friedman.
Site includes R/S+ Code
- The best multivariate analysis and Statistical
Learning textbook I know of; web site includes software. From a modern
and sophisticated computational statistics viewpoint, but quite readable.
Compares methods from trees to neural nets,
kernel methods, and support vector machines,
though nothing on genetic algorithms. You can even learn the meaning of useful
things like bootstrapping and boosting and other post-1960's statistical jargon!
- a massive toolkit, including visualization, feature selection,
many evaluation tools, including calibration curves and ROC
(Receiver Operating Characteristics = efficiency for signal vs fraction of
background: true positives vs. false positives). Practically all major algorithms
from machine learning. Python is a popular interface to this library.
Multivariate Analysis Software (some older items, but useful)
- libsvm, SVMlight, and PRTools are popular Pattern Recognition software (thanks: MSU Computer Science)
- support-vectormachines.org has more svm software and information
MIT Classification Software collection--easy to compare methods
mixture of commercial and academic software links
- Machine Learning Resources online
- Neural Network Software list, see also SNNS
popular in Babar
- ROC curves :
- Note: external criteria
define the best efficiency point to select, and that often no single algorithm
dominates at all efficiencies. There is a considerable gap
between the machine learning and statistics communities, which Elements
of Statistical Learning by Hastie et al tries to bridge.
- weka is one Java toolkit
- mloss is a sortable machine learning repository of free resources; also try wiki machine learning
Most common languages are Python and R
- A Python data analysis distribution should include SciPy NumPy and MatplotLib
- scikit-learn powerful machine learning Python package, and very good summary of the different methods
- Theano; a very powerful (Linux) library for Python
http://pandas.pydata.org/ brings dataframes to Python
caret most comprehensive predictive modeling package in R; book: http://appliedpredictivemodeling.com/
Rattle, a GUI for data mining in R; book: http://goo.gl/kaUqr2
- Plotly Powerful visualization package
Julia interesting new language for data analysis (getting closer to maturity).
An Introduction to Statistical Learning: http://www-bcf.usc.edu/~gareth/ISL/ R based; simple than Hastie et al
Practical Data Science with R
- RDataMining data mining examples in R
- http://deeplearning.net/tutorial/ recent fad in machine learning
- https://www.coursera.org/specializations/jhudatascience Data Science Specialization from Johns Hopkins
- http://cs109.org/<http://cs109.org/readings.php> Harvard data science class.
- blogs: yhatq, wesmckiney, hillarymason, r-bloggers
- http://www.quandl.com/ datasets
Google rankings, and Glen Cowan's and Eric Feiglson's pages got me started.
The following people (and some others I've forgotten) have provided me with
several useful links as well as some excellent suggestions which I have unaccountably
Tom Loredo (Cornell Astronomy); Rene Brun (CERN); Paul Padley (Rice); Jim Kowalkowski
(Fermilab) John Rice (Berkeley Statistics); Louis Lyons (Oxford/UCL); Ilya Narsky
(Matlab), Deb Davis (statistics teacher); Yannis Katsnos, Aous Abdo, Sarah G. Williams (data miners)
Last updated November 2015