Thank you Soutrik

It's quite expensive. I'm not sure yet how much I'm going to use it. What do you think about Netica? They have this limited free version.

Despite your hesitation about command line programs, I would still recommend R or WinBUGs since both of the programs have GUI "accessories" that make them more approachable. I personally like R because it is easily configurable and you can tell it exactly what you want it to do by writing your own functions.

For WinBUGs, I would direct you to BugsXLA, an Excel add-in that allows you to perform Bayesian analysis. I have not personally used it, but I have heard good things.

For R, there are programs like Tinn-R & R-Commander that add a GUI to R. I would also highly recommend Kruschke's "Doing Bayesian Analysis" text as a reference for the Bayesian novice. It's very approachable and provides all the code to do what you need. And it has puppies on the cover. Everyone loves puppies.

Finally, I would also recommend the UCD's BEST website (http://www.epi.ucdavis.edu/diagnostictests/index.html) if you are doing diagnostic test evaluation. There is a code repository and a very handy beta distribution generator called "BetaBuster" for download.

Best of luck!

Thank you William, I will give a try to all of them. BugsXLA sounds right. I like the idea about Excel addon - very handy when you are writing a text with HQ-figures. An thank you for the book tip. I already ordered "Risk Assessment and Decision Analysis with Bayesian" and "Data Analysis: A Bayesian Tutorial". But I think I'll also try the book you recommended.

Soutrik, here is a link to Netica - http://www.norsys.com/netica.html. And, unfortunately, I don't have any collaborator in UK.

OpenBugs

I ordered the Kruschke's book after I read the first chapter I found on internet. I installed WinBUGS and BugsXLA. So I'll be preoccupied for few weeks. Thanks for help.

I use Matlab to write code to do Bayesian computation.

I think, I'm going to read the Kruschke's book and make all exercises he suggested. So, it's R and Bugs for now. Actually, R doesn't seem bad at all. It's just been almost ten years since I wrote a code.

Late to this. Just wanted to mention that netica is designed for Bayesian Belief Networks whereas BUGS, JAGS, etc are generally for Bayesian statistical models. They all use Bayes theorem of course, and are very similar (I guess you could build a BBN in BUGS), but I would say that netica is not a GUI solution for Bayesian statistical modelling per se.

Good point, Ben. I didn't realize it.

FirstBayes (www.firstbayes.co.uk) is made to be basic and introductory.

Fausto: the solution to your challenge is that the probability is the integral over the posterior of mu from 100 to +inf (or to the end of the domain of mu).

The posterior is the probability distribution over possible values of mu, given the data. It is obtained through relatively standard probability calculus (using Bayes' theorem) and requires a specification of both prior and likelihood. Those specifications depend on the particular situation.

The posterior is obtained as the product of the prior and the likelihood, the specification of which requires some knowledge about the content area. You seem to be asking a poorly-specified question. A sensible prior will depend on whether the data are temperatures, life spans, scores on a test, etc. From the information you provide, the best answer you can get is going to be in terms of unspecified priors and likelihoods. Given those, the probability you seek is the mass of the posterior right of 100.

I think a fundamental idea you are not appreciating is that any method that provides a well-specified solution to a poorly-specified problem must necessarily be flawed. I know your problem was intentionally poorly specified and that is why the solution is only specified up to that degree. Now that you've added some information (that the data are life spans), we can construct a prior that captures the distribution of mean life spans. I'm not going to do that because I don't know anything about actuarial statistics, but one could.

The mean is going to be greater than 100 with some probability, which is obtained through integrating the posterior of the mean over the appropriate range, and which you can then (at your discretion) use to decide on your investment.

Well, yes, if you change the specifics of the problem, the solution may indeed change. That is not a weakness of Bayesian methods, that is a property of the universe.

There are infinitely many ways to come up with a point estimate of a mean. If you are still interested in the probability that the mean is in some range (e.g., greater than 100), that probability is in fact obtained by integrating its posterior over that range.

The qualities of Bayesian methods are very neatly described in E.T. Jaynes' excellent book "Probability theory: The logic of science", which I am willing to wager you would enjoy.

I will try and answer the original question!

If you want a free book to see what modern Bayesian analysis can achieve and learn something along the way of how it works then try Bill Browne's MCMC estimation in MLwin which does indeed have a GUI interface and provides lots of non-trivial examples:

Browne, W.J. (2012) MCMC Estimation in MLwiN, v2.26. Centre for Multilevel Modelling, University of Bristol.

http://www.bristol.ac.uk/cmm/software/mlwin/download/2-26/mcmc-print.pdf

I suggest GeNIe from University of Pittsburgh. Very easy to use, feature-rich and free! The supporting documentation is good too.

http://genie.sis.pitt.edu/

I suggest the R2STATS GUI, and the book by Y. Noël from the the Springer Use R! series.

The only problem is that for now the book is only available in French (and I know it is a pitty !!)

I have bluffed my way into MCMC using t-walk (I had no previous background)

http://www.cimat.mx/~jac/twalk/

Its really worth taking it for a test drive!

I would suggest EPIDAT (it can be download from http://dxsp.sergas.es/ ). At present, it is only in Spanish, but an English version will be avaiblable soon. 

I use GeNIe/SMILE (University of Pittsburgh).

SMILE is a fully platform independent library of C++ classes implementing graphical probabilistic and decision-theoretic models, such as Bayesian networks. SMILE is released as a dynamic link library (DLL). There are also several SMILE wrappers, such as SMILE.NET (.NET interface - very good), jSMILE (Java interface) etc. Versions available for Windows, Unix (Solaris), Linux, Mac etc.

GeNIe environment for the creation of decision theoretic models. Main features are graphical editor to create and modify network models and the SMILE Engine. You may develop models in GeNIe and create a custom interface for them using SMILE. 

System have thorough and good documentation.

http://genie.sis.pitt.edu/

May be too late now. I am a Dentist who self taught these things for my own projects. It's a steep learning curve but not unachievable.

I saw that you started with R and BUGS.

An upgrade at least for speed, and ease to debug your model would be moving to Stan http://mc-stan.org/.

Stan is opensource, Bayesian inference engine which does Hamiltonial Monte Carlo with No-U-turn algorithm.

I believe what one needs to self teach these methods is an excellent community to help you. There is a ton of tutorials and case studies in their website with an excellent discussion forum (http://discourse.mc-stan.org/)

were fellow members are happy to review your codes!.

If you are running it from R/RStudio, there is two excellent packages -

rstanarm - https://cran.r-project.org/web/packages/rstanarm/vignettes/rstanarm.html

and brms - https://cran.r-project.org/web/packages/brms/index.html

You can perform almost all common (and some real cool ones too) using them with out any need to write model codes!

SpiceLogic Bayesian doctor is a nice GUI software that may suit your need. It is very intuitive and simple enough to be used for beginners. https://www.spicelogic.com/Products/Bayesian-network-software-20

You can easily model Bayesian Network or Bayesian Inference, belief update upon evidences etc.

It is not expensive either.. affordable student price.

Hi, I'm  a beginner user of GeNiE/Smile and i need your help about an issue in Learning Parameters Phase !

How can i generate the CPT from CSV file containing the values of each node in 89 records ? I want to calculate the probability of each node's state from data and not to fill in the CPT manually as indicated in the documentation .

Salsabill Bel Hadj Sgayer

I would recommend asking this as a separate question not added to an old thread.

Dear Kelvyn Jones ,

It's done ^^

I think SPSS and Minitab are better for a beginner.

What about Eviews 8 and higher ones?