dkane
October 14, 2020, 11:55pm
1
rstanarm::posterior_predict() creates objects of class "ppd" "matrix" "array" . I want to convert such objects to clean tibbles. I tried:
library(tidyverse)
library(rstanarm)
#> Loading required package: Rcpp
#> This is rstanarm version 2.21.1
#> - See https://mc-stan.org/rstanarm/articles/priors for changes to default priors!
#> - Default priors may change, so it's safest to specify priors, even if equivalent to the defaults.
#> - For execution on a local, multicore CPU with excess RAM we recommend calling
#> options(mc.cores = parallel::detectCores())
obj <- stan_glm(data = women, height ~ 1, refresh = 0)
pp <- posterior_predict(obj)
pp %>%
as_tibble()
#> # A tibble: 4,000 x 15
#> `1` `2` `3` `4` `5` `6` `7` `8` `9` `10` `11` `12` `13`
#> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd>
#> 1 66.4… 61.5… 67.9… 59.4… 64.8… 60.4… 60.5… 63.9… 67.1… 66.1… 66.6… 65.7… 67.0…
#> 2 70.3… 60.0… 57.9… 58.2… 67.6… 63.1… 52.7… 62.5… 67.3… 51.3… 59.9… 63.7… 56.9…
#> 3 65.4… 65.1… 59.6… 69.3… 73.6… 63.8… 69.2… 72.8… 71.9… 61.6… 73.8… 64.0… 68.1…
#> 4 67.3… 63.7… 67.9… 66.5… 60.1… 63.1… 61.1… 71.1… 71.9… 63.7… 64.2… 61.4… 70.9…
#> 5 66.3… 62.2… 61.1… 66.5… 63.3… 68.3… 61.2… 59.3… 63.4… 69.5… 65.3… 61.0… 67.4…
#> 6 77.7… 55.4… 64.5… 67.2… 72.8… 55.1… 60.2… 67.3… 53.8… 59.5… 67.8… 65.6… 53.0…
#> 7 67.4… 68.8… 68.3… 70.9… 67.4… 65.4… 66.5… 64.7… 65.1… 69.9… 60.5… 68.3… 60.9…
#> 8 63.6… 81.7… 73.0… 61.7… 69.3… 69.9… 68.7… 69.4… 64.8… 62.0… 61.5… 65.3… 65.0…
#> 9 58.8… 63.4… 64.5… 67.5… 64.9… 60.1… 66.8… 65.6… 62.5… 63.8… 65.4… 58.5… 63.0…
#> 10 62.5… 60.3… 70.7… 72.0… 72.2… 76.3… 71.4… 65.9… 66.6… 72.3… 67.4… 64.1… 68.5…
#> # … with 3,990 more rows, and 2 more variables: `14` <ppd>, `15` <ppd>
Created on 2020-10-14 by the reprex package (v0.3.0)
I would like each column to be a double, as we would get if pp were a simple matrix. But, as you can see, each column is, itself, an object of class ppd. How can I convert an object of class "ppd" "matrix" "array" to a clean tibble?
For converting to a tibble, I believe the tidybayes package has you covered (see e.g., https://cran.r-project.org/web/packages/tidybayes/vignettes/tidy-rstanarm.html#posterior-fits )
Also, you can use as.matrix on the results of posterior_predict to convert to matrix.
Does that answer your question?
dkane
October 15, 2020, 5:44pm
3
Thanks for your help! But adding as.matrix() does not solve the problem. The resulting object still has "ppd" as the type for each column when what I want is a simple dbl.
library(tidyverse)
library(rstanarm)
#> Loading required package: Rcpp
#> This is rstanarm version 2.21.1
#> - See https://mc-stan.org/rstanarm/articles/priors for changes to default priors!
#> - Default priors may change, so it's safest to specify priors, even if equivalent to the defaults.
#> - For execution on a local, multicore CPU with excess RAM we recommend calling
#> options(mc.cores = parallel::detectCores())
obj <- stan_glm(data = women, height ~ 1, refresh = 0)
pp <- posterior_predict(obj)
pp %>%
as.matrix() %>%
as_tibble()
#> # A tibble: 4,000 x 15
#> `1` `2` `3` `4` `5` `6` `7` `8` `9` `10` `11` `12` `13`
#> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd> <ppd>
#> 1 63.5… 70.2… 64.4… 55.8… 62.3… 65.7… 57.5… 63.3… 65.4… 58.5… 68.0… 61.8… 60.4…
#> 2 58.1… 55.8… 70.7… 70.0… 69.1… 73.1… 68.9… 51.4… 75.9… 69.2… 76.1… 69.8… 68.9…
#> 3 68.3… 69.4… 64.0… 57.3… 65.2… 64.5… 61.3… 64.4… 76.7… 64.9… 61.0… 62.0… 63.9…
#> 4 66.1… 71.4… 62.1… 59.0… 61.3… 61.0… 53.2… 72.4… 64.6… 64.4… 66.9… 62.0… 71.6…
#> 5 70.7… 63.4… 67.2… 63.3… 65.6… 65.6… 67.9… 63.3… 63.5… 62.1… 59.7… 61.3… 71.5…
#> 6 63.9… 52.4… 71.5… 59.7… 67.0… 69.1… 74.2… 64.5… 60.9… 58.2… 66.1… 69.6… 66.2…
#> 7 67.6… 61.0… 64.8… 64.9… 69.0… 72.3… 64.0… 72.5… 70.4… 71.8… 70.1… 66.6… 68.1…
#> 8 63.6… 74.4… 70.6… 62.3… 63.0… 65.9… 69.4… 75.7… 65.2… 58.5… 69.9… 62.6… 65.8…
#> 9 68.7… 65.7… 66.4… 62.7… 67.5… 63.0… 73.9… 62.9… 60.5… 65.5… 67.4… 70.6… 56.6…
#> 10 65.9… 60.5… 62.3… 67.9… 69.1… 66.8… 72.5… 67.6… 55.8… 66.3… 62.4… 60.3… 61.4…
#> # … with 3,990 more rows, and 2 more variables: `14` <ppd>, `15` <ppd>
Created on 2020-10-15 by the reprex package (v0.3.0)
And thanks for the link to tidybayes, but the section you link to is for doing the equivalent of rstanarm::posterior_linpred(), when I am creating poster_predict() objects.
dkane
October 30, 2020, 2:29am
4
Here is an option that seems to work:
library(tidyverse)
library(rstanarm)
#> Loading required package: Rcpp
#> This is rstanarm version 2.21.1
#> - See https://mc-stan.org/rstanarm/articles/priors for changes to default priors!
#> - Default priors may change, so it's safest to specify priors, even if equivalent to the defaults.
#> - For execution on a local, multicore CPU with excess RAM we recommend calling
#> options(mc.cores = parallel::detectCores())
obj <- stan_glm(data = women, height ~ 1, refresh = 0)
posterior_predict(obj) %>%
as_tibble() %>%
mutate(across(everything(), as.numeric))
#> # A tibble: 4,000 x 15
#> `1` `2` `3` `4` `5` `6` `7` `8` `9` `10` `11` `12` `13`
#> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 58.9 63.9 49.2 58.6 75.0 62.4 67.4 51.4 62.8 67.1 67.7 57.8 54.2
#> 2 75.6 65.2 63.5 65.7 60.1 63.6 74.4 57.3 55.6 60.2 61.9 60.7 57.2
#> 3 56.5 64.8 65.0 52.7 59.9 70.0 64.6 57.9 70.8 68.5 75.0 49.7 62.3
#> 4 50.0 64.3 73.4 56.4 58.2 70.3 57.9 59.1 57.0 65.3 54.8 55.1 69.8
#> 5 64.1 62.0 64.4 58.8 58.6 62.3 63.9 65.2 62.2 66.7 61.7 59.4 70.4
#> 6 70.6 67.3 63.6 68.3 55.1 62.5 61.9 70.6 68.9 65.9 67.1 69.4 61.6
#> 7 62.8 65.9 63.7 69.3 65.1 67.1 67.0 66.9 71.9 75.2 69.7 67.5 65.2
#> 8 58.6 69.8 65.8 67.2 66.4 62.0 64.9 68.2 78.3 67.9 70.4 71.7 57.8
#> 9 65.5 66.4 68.0 68.1 63.4 62.0 68.8 61.2 67.6 71.3 70.0 67.2 68.2
#> 10 67.4 66.7 64.9 64.9 68.1 60.2 59.9 64.9 62.0 58.4 64.6 70.4 65.7
#> # … with 3,990 more rows, and 2 more variables: `14` <dbl>, `15` <dbl>
Created on 2020-10-29 by the reprex package (v0.3.0)
I hate it because, for teaching purposes, that mutate(across nonsense is very hard for students to grasp. Are there better solutions?
dkane
October 30, 2020, 12:48pm
6
Thanks! That is much better!
system
November 20, 2020, 12:48pm
7
This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.