`R/recover_types.R`

`recover_types.Rd`

Decorate a Bayesian model fit or a sample from it with types for
variable and dimension data types. Meant to be used before calling
`spread_draws`

or `gather_draws`

so that the values returned by
those functions are translated back into useful data types.

recover_types(model, ...)

model | A supported Bayesian model fit. Tidybayes supports a variety of model objects; for a full list of supported models, see tidybayes-models. |
---|---|

... | Lists (or data frames) providing data prototypes used to convert
columns returned by |

A decorated version of `model`

.

Each argument in `...`

specifies a list or data.frame. The `model`

is decorated with a list of constructors that can convert a numeric column
into the data types in the lists in `...`

.

Then, when `spread_draws`

or `gather_draws`

is called on the decorated
`model`

, each list entry with the same name as the variable or a dimension
in `variable_spec`

is a used as a prototype for that variable or dimension ---
i.e., its type is taken to be the expected type of that variable or dimension.
Those types are used to translate numeric values of variables back into
useful values (for example, levels of a factor).

The most common use of `recover_types`

is to automatically translate
dimensions of a variable that correspond to levels of a factor in the original data back into
levels of that factor. The simplest way to do this is to pass in the data
frame from which the original data came.

Supported types of prototypes are factor, ordered, and logical. For example:

if

`prototypes$v`

is a factor, the v column in the returned draws is translated into a factor using`factor(v, labels=levels(prototypes$v), ordered=is.ordered(prototypes$v))`

.if

`prototypes$v`

is a logical, the v column is translated into a logical using`as.logical(v)`

.

Additional data types can be supported by providing a custom implementation
of the generic function `as_constructor`

.

library(dplyr) library(magrittr) if(require("rstan", quietly = TRUE)) { # Here's an example dataset with a categorical predictor (`condition`) with several levels: set.seed(5) n = 10 n_condition = 5 ABC = data_frame( condition = rep(c("A","B","C","D","E"), n), response = rnorm(n * 5, c(0,1,2,1,-1), 0.5) ) # We'll fit the following model to it: stan_code = " data { int<lower=1> n; int<lower=1> n_condition; int<lower=1, upper=n_condition> condition[n]; real response[n]; } parameters { real overall_mean; vector[n_condition] condition_zoffset; real<lower=0> response_sd; real<lower=0> condition_mean_sd; } transformed parameters { vector[n_condition] condition_mean; condition_mean = overall_mean + condition_zoffset * condition_mean_sd; } model { response_sd ~ cauchy(0, 1); // => half-cauchy(0, 1) condition_mean_sd ~ cauchy(0, 1); // => half-cauchy(0, 1) overall_mean ~ normal(0, 5); //=> condition_mean ~ normal(overall_mean, condition_mean_sd) condition_zoffset ~ normal(0, 1); for (i in 1:n) { response[i] ~ normal(condition_mean[condition[i]], response_sd); } } " m = stan(model_code = stan_code, data = compose_data(ABC), control = list(adapt_delta=0.99), # 1 chain / few iterations just so example runs quickly # do not use in practice chains = 1, iter = 500) # without using recover_types(), the `condition` column returned by spread_draws() # will be an integer: m %>% spread_draws(condition_mean[condition]) %>% median_qi() # If we apply recover_types() first, subsequent calls to other tidybayes functions will # automatically back-convert factors so that they are labeled with their original levels # (assuming the same name is used) m %<>% recover_types(ABC) # now the `condition` column with be a factor with levels "A", "B", "C", ... m %>% spread_draws(condition_mean[condition]) %>% median_qi() }#>#>#> #> #>#> #>#>#> #>#>#> #>#>#> #>#> #> SAMPLING FOR MODEL 'cd9b191279da85ce85fc8454c66a58e5' NOW (CHAIN 1). #> Chain 1: Gradient evaluation took 0 seconds #> Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 0 seconds. #> Chain 1: Iteration: 1 / 500 [ 0%] (Warmup) #> Chain 1: Iteration: 50 / 500 [ 10%] (Warmup) #> Chain 1: Iteration: 100 / 500 [ 20%] (Warmup) #> Chain 1: Iteration: 150 / 500 [ 30%] (Warmup) #> Chain 1: Iteration: 200 / 500 [ 40%] (Warmup) #> Chain 1: Iteration: 250 / 500 [ 50%] (Warmup) #> Chain 1: Iteration: 251 / 500 [ 50%] (Sampling) #> Chain 1: Iteration: 300 / 500 [ 60%] (Sampling) #> Chain 1: Iteration: 350 / 500 [ 70%] (Sampling) #> Chain 1: Iteration: 400 / 500 [ 80%] (Sampling) #> Chain 1: Iteration: 450 / 500 [ 90%] (Sampling) #> Chain 1: Iteration: 500 / 500 [100%] (Sampling) #> Chain 1: Elapsed Time: 0.103 seconds (Warm-up) #> Chain 1: 0.108 seconds (Sampling) #> Chain 1: 0.211 seconds (Total)#> # A tibble: 5 x 7 #> condition condition_mean .lower .upper .width .point .interval #> <chr> <dbl> <dbl> <dbl> <dbl> <chr> <chr> #> 1 A 0.186 -0.0838 0.515 0.95 median qi #> 2 B 1.00 0.664 1.32 0.95 median qi #> 3 C 1.83 1.54 2.16 0.95 median qi #> 4 D 1.03 0.652 1.35 0.95 median qi #> 5 E -0.917 -1.18 -0.577 0.95 median qi