support of brms regression models #89
Conversation
|
I have added basic support of brmrs model. It still needs to be tested with different cases. I had to deal with the fact that there is no Please have a look in particular to the folowing files:
In addition, library(broom.helpers)
library(broom.mixed)
#> Registered S3 method overwritten by 'broom.mixed':
#> method from
#> tidy.gamlss broom
library(dplyr)
#>
#> Attachement du package : 'dplyr'
#> The following objects are masked from 'package:stats':
#>
#> filter, lag
#> The following objects are masked from 'package:base':
#>
#> intersect, setdiff, setequal, union
load(system.file("extdata", "brms_example.rda", package="broom.mixed"))
brms_crossedRE %>% tidy_plus_plus(intercept = T) %>% knitr::kable()
library(brms)
#> Le chargement a nécessité le package : Rcpp
#> Loading 'brms' package (version 2.14.4). Useful instructions
#> can be found by typing help('brms'). A more detailed introduction
#> to the package is available through vignette('brms_overview').
#>
#> Attachement du package : 'brms'
#> The following object is masked from 'package:stats':
#>
#> ar
mtcars <- mtcars %>% mutate(gear = gear %>% factor, vs = vs %>% factor, carb = carb %>% factor)
contrasts(mtcars$vs) <- "contr.sum"
model_brms_2 <- brm(mpg ~ (1 | gear) + hp + vs + carb, data = mtcars, backend = "cmdstanr", refresh = 0)
#> Compiling Stan program...
#> Start sampling
#> Running MCMC with 4 sequential chains...
#>
#> Chain 1 finished in 0.9 seconds.
#> Chain 2 finished in 0.9 seconds.
#> Chain 3 finished in 0.9 seconds.
#> Chain 4 finished in 0.9 seconds.
#>
#> All 4 chains finished successfully.
#> Mean chain execution time: 0.9 seconds.
#> Total execution time: 4.0 seconds.
#>
#> Warning: 4 of 4000 (0.0%) transitions ended with a divergence.
#> This may indicate insufficient exploration of the posterior distribution.
#> Possible remedies include:
#> * Increasing adapt_delta closer to 1 (default is 0.8)
#> * Reparameterizing the model (e.g. using a non-centered parameterization)
#> * Using informative or weakly informative prior distributions
model_brms_2 %>% tidy_plus_plus(intercept = T) %>% knitr::kable()
Created on 2021-01-27 by the reprex package (v0.3.0) |
|
Regarding:
Family: gaussian
Links: mu = identity; sigma = identity
Formula: mpg ~ wt + (1 | cyl) + (1 + wt | gear)
Data: mtcars (Number of observations: 32)
Samples: 2 chains, each with iter = 100; warmup = 50; thin = 1;
total post-warmup samples = 100
Group-Level Effects:
~cyl (Number of levels: 3)
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sd(Intercept) 4.57 3.64 0.35 14.63 1.12 13 50
~gear (Number of levels: 3)
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sd(Intercept) 4.57 3.21 0.06 12.84 1.00 43 48
sd(wt) 1.71 1.32 0.12 4.92 1.03 27 77
cor(Intercept,wt) -0.29 0.58 -0.96 0.89 1.00 45 61
Population-Level Effects:
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
Intercept 33.89 3.81 26.28 40.47 1.04 54 31
wt -3.91 1.39 -6.27 -1.17 1.03 27 60
Family Specific Parameters:
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sigma 2.59 0.41 2.03 3.63 1.01 118 77
Samples were drawn using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1). |
|
@larmarange thank you kindly for updating to accommodate brms models. I see that it took some re-structuring and adding new S3 functions. |
|
Hey guys, I don't know if you are aware but library(brms)
fit <- brm(bf(mpg ~ 1 + (1 + hp | cyl), decomp = "QR"), data = mtcars, refresh = 0)
broom.mixed::tidy(fit, conf.int = TRUE)r$> broom.mixed::tidy(fit, conf.int = TRUE)
Registered S3 method overwritten by 'broom.mixed':
method from
tidy.gamlss broom
# A tibble: 5 x 8
effect component group term estimate std.error conf.low conf.high
<chr> <chr> <chr> <chr> <dbl> <dbl> <dbl> <dbl>
1 fixed cond NA (Intercept) 20.8 3.52 13.9 28.0
2 ran_pars cond cyl sd__(Intercept) 7.09 4.12 1.21 17.5
3 ran_pars cond cyl sd__hp 0.0440 0.0360 0.00184 0.143
4 ran_pars cond cyl cor__(Intercept).hp -0.221 0.554 -0.976 0.881
5 ran_pars cond Residual sd__Observation 3.27 0.480 2.48 4.33 PS: Sorry |
|
Dear @storopoli See also #90 about better management of mixed models in general. |
new var_type "ran_pars"
|
@ddsjoberg R CMD check fails for R3.5 and R3.4. It says that "brms" is not available, although installed through historical version of CRAN. Any idea? |
|
i'll take a look tomorrow. I may post a bunch of commits to test FYI |
Maybe an option could be to not force suggested packages for R 3.4 and 3.5? |
Do you mean not do checks on 3.5 and 3.4, or is there a way to skip the suggested packages? I looked at 3.4 and it seems to be an issue installing rstan. Either of the options above seem reasonable |
I think you may find that R Cmd check will fail if every package in Suggests cannot be installed (even if the checks using that package are skipped when it's not installed). |
|
OK with the change all checks passed. Please note that for R 3.4, some tests are skipped due to the fact that brms is not available. However, I guess that it is acceptable for R 3.4 and R 3.5 to skip corresponding tests if some packages are not available. For R release, oldrel and devel, we still force all suggested packages to pass the tests. |
|
That is amazing! |
|
I have separated checks in 2 workflows. For R 3.4 and R 3.5 I have added |
Codecov Report
@@ Coverage Diff @@
## master #89 +/- ##
==========================================
- Coverage 99.85% 99.71% -0.14%
==========================================
Files 33 34 +1
Lines 1375 1414 +39
==========================================
+ Hits 1373 1410 +37
- Misses 2 4 +2
|
fix #87
and basic support for crr() models
fix #91