Dmenne BUG #29
Dmenne BUG #29
Conversation
As of Aug 4, 2015 ./Ch.10/sesame_one_pred_2b.stan ./Ch.10/sesame_one_pred_b.stan ./Ch.12/radon_group_chr.stan ./Ch.12/radon_intercept_chr.stan ./Ch.12/radon_no_pool_chr.stan ./Ch.13/13.4_UnderstandingCorrelationsBetweenIntercepts&Slopes.R ./Ch.13/earnings_latin_square_chr.stan ./Ch.13/earnings_vary_si_chr.stan ./Ch.13/pilots_chr.stan ./Ch.13/radon_inter_vary_chr.stan ./Ch.13/radon_vary_si_chr.stan ./Ch.14/pilots_chr.stan ./Ch.16/16.3_FittingAndUnderstangingAVaryingInterceptMultilevelModelUsingRStan.R ./Ch.17/17.2_radon_multi_varying_coef.stan ./Ch.17/17.5_multilevel_poisson.stan ./Ch.17/17.6_multilevel_ordered_categorical.stan ./Ch.17/17.7_latent_glm.stan ./Ch.17/17.7_robit.stan ./Ch.19/item_response.stan ./Ch.19/radon_chr.stan ./Ch.19/radon_redundant_chr.stan ./Ch.20/hiv_chr.stan ./Ch.20/hiv_inter_chr.stan ./Ch.21/finite_populations.stan ./Ch.21/multiple_comparison.stan ./Ch.21/r_sqr.stan ./Ch.21/radon_vary_intercept_floor.stan ./Ch.21/radon_vary_intercept_floor_chr.stan ./Ch.21/radon_vary_intercept_floor2_chr.stan ./Ch.21/radon_vary_intercept_nofloor_chr.stan ./Ch.22/anova_radon_nopred_chr.stan ./Ch.23/educational_subsidy.stan ./Ch.23/electric.stan ./Ch.23/electric_1a_chr.stan ./Ch.23/electric_1b_chr.stan ./Ch.23/electric_1c_chr.stan ./Ch.23/electric_chr.stan ./Ch.24/dogs_check.stan ./Ch.24/dogs_log.stan ./Ch.3/kidscore_momhs.stan ./Ch.4/earn_height.stan ./Ch.8/8.2_FakeDataSimulationToUnderstandResidualPlots.R
The suggested correction in http://stats.stackexchange.com/questions/143359/when-is-mu-a-used-in-this-stan-example is not correct. Changed variable names to those used in equation 13.9, i.e. gamma and delta instead of a and b. Removed unused mu_a and mu_b, and added common mu to really force gamma and delta into approx N(0,).
Add rats_stanified, with flat priors and avoiding sigma-squared detours. Print mean and stddev for all models, BUGS and Gelfand (original citation).
Check random default initialization Compare with BUGS result in manual
Add simplified version that does not use tau but estimates sigma directly. Narrower alpha priors (from 1000 to 1)
…files, will be corrected in bulk later.
Add stanified version
…UGS manual. Add stanified version
Merge branch 'dmenne-dev-bugs' of https://github.com/dmenne/example-models into dmenne-dev-bugs Conflicts: bugs_examples/vol1/surgical/surgical.R bugs_examples/vol1/surgical/surgical_stanified.stan
|
Hi, I'm also going through the example models in BUGS. Is it worth me continuing on volume 2 so as not to duplicate the workload? Best, Ben
|
|
Hi, Ben, I do this for my own education, but then Bob asked me to submit a pull request. It certainly would be good to coordinate. What have you done already? Dividing by the volume line seems ok, but I don’t mind trying other examples. Dieter Von: benlambertStan [mailto:notifications@github.com] Hi, I'm also going through the example models in BUGS. Is it worth me continuing on volume 2 so as not to duplicate the workload? Best, Ben
— |
|
Hi Dieter, I've also gone through most of the examples you mention, as well as the dogs and beta blockers examples from volume 1. Will commit my models to the GitHub next week sometime. I'll start to look at volume 2 examples now. Cheers, Ben
|
|
@benlambertStan, I would suggest you put what you have on a public branch now rather than wait another week to do so. Since we have a little momentum, it would be great if you coordinated and compared models. I'd be interested in what both of you are trying to do in terms of having models run. This is related to #26. |
|
and thanks, @dmenne. @benlambertStan, if you've already looked at this pull request and think everything is reasonable, either write that as a comment or feel free to merge into the branch. If you don't get to it, I'll do it soon. |
|
Yes, will do - it will be good to compare across both of our runs.
|
|
I'll second what Daniel said --- it's better to get things into Thanks.
P.S. We're about to figure out how to add developer names to all the
|
|
@syclik I see your point in #26: it would be good to have TravisCI run these examples as a testbed. It was not the priority for me, my own interest is in using these examples to get rid of BUG-habits. The examples are a bit too close to the BUGs version, which is fine for a starter. I am also not fluent in makefiles, using interactive mode most of the time My suggestion: please modify one of our examples so it fits your requirements. Adding testthat or whatever? Then I could try to follow that path. I also had considered to create the examples in Notebook format (http://rmarkdown.rstudio.com/r_notebook_format.html) so that comments could be inserted. |
|
@syclik As a separate point: the links to bug on top of each file are still incorrect. I wanted to correct these globally, but it would mean that all files, even the non-revised ones, will be changed and need to be merged. What's the recommended procedure to follow? |
|
Pull request. Merge early and often with the develop (or master)
|
…nto dmenne-dev-bugs
@dmenne fixed a lot of the bugs examples.
-- using vectorized if possible
-- replace 1/sigmasquared contortions by sigma
-- delete duplicate or non-required files
-- somewhat narrower or wider (=default) priors
-- test if program converges by running with default random init instead of the constant init.
To discuss: there is a warning to increment in old version "rats_vec_unit.stan". Result seem ok, and I do not understand what to do; I tend to remove the example, it is not very useful, but try to understand first.