Thanks T. Once you have your learning or reference table made of model index x parameter x statistics, you can call the down-the-shelf random forest algorithm to explain the model index by the statistics. Once the algorithm has built the forest you can (a) enter the observed statistics which will return a most frequent model index, which will be the estimated model and (b) derive the simulated statistics closest to the observed ones, i.e the most often in the same leaf of the trees of the forest, which will give you an ABC posterior sample, based on the random forest. I hope this helps. X

]]>I am not a statistician nor a mathematician, just a biologist interested in ABC for practical issues, and I feel like I did wrong until now …

I am convinced that random forests are more relevant and less arbitrary than previous approaches for model choice.

However, I have a stupid question … Once I have all of my simulations, meaning, a huge array of (iterations)x(statistics), how can I apply to it the good practices you recommend?

Sorry for the very pragmatic question.

All the best, and congrates again for your contribution to ABC.

T. ]]>

Thanks Dennis. Explanation: I wrote this post before the paper was arXived. And the arXiv team no longer provides the arXiv number until acceptance. Hence put the link to new postings… Feel free to write a guest post, by the way!!

]]>Full circle??? Not really as I started from decision theory with strong Bayesian inclinations. This move is induced by the realisation we cannot currently produce a trustworthy approximation to posterior probabilities, while a machine-learning tool can rank models with (a) some degree of efficiency and (b) still a Bayesian flavour. Plus, I am having more and more qualms about using posterior probabilities per se. Thanks for pointing out the typo. And you are welcome to a guest post if you feel like it. Best wishes for ABC in Syndey!

]]>Does this mean you have now come full circle, Christian? :-)

]]>