Feature tidyselect

[jgabry]

Closes #161
Closes #183

This PR allows the pars argument to MCMC plots be a list of quosures, as returned by dplyr::vars(). Internally bayesplot will call tidyselect::vars_select() if pars is a list of quosures. If pars is just a character vector then the behavior is unchanged from previous versions of bayesplot, which preserves backwards compatibility.

Example usage

x <- example_mcmc_draws()
mcmc_hist(x, pars = vars(-contains("beta"))
mcmc_hist(x, pars = vars(alpha, sigma))

Important changes for review

The important changes for review are in helpers-mcmc.R

bayesplot/R/helpers-mcmc.R

Lines 31 to 38 in eef7bc8

	if (rlang::is_quosures(pars)) {
	pars <- tidyselect_parameters(complete_pars = parameter_names(x),
	pars_list = pars)
	} else {
	pars <- select_parameters(complete_pars = parameter_names(x),
	explicit = pars,
	patterns = regex_pars)
	}

and the new file tidy-params.R, which is mostly for examples of tidy parameter selection but also contains this new internal function:

bayesplot/R/tidy-params.R

Lines 73 to 81 in eef7bc8

	tidyselect_parameters <- function(complete_pars, pars_list) {
	helpers <- tidyselect::vars_select_helpers
	pars_list <- lapply(pars_list, rlang::env_bury, !!! helpers)
	selected <- tidyselect::vars_select(.vars = complete_pars, !!! pars_list)
	if (!length(selected)) {
	stop("No parameters were found matching those names.", call. = FALSE)
	}
	return(unname(selected))
	}

[codecov-io]

Codecov Report

Merging #188 into master will increase coverage by 0.02%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #188      +/-   ##
==========================================
+ Coverage   99.33%   99.35%   +0.02%     
==========================================
  Files          30       31       +1     
  Lines        4212     4502     +290     
==========================================
+ Hits         4184     4473     +289     
- Misses         28       29       +1

Impacted Files	Coverage Δ
R/helpers-shared.R	100% <ø> (ø)	⬆️
R/mcmc-scatterplots.R	99.24% <ø> (-0.21%)	⬇️
R/helpers-mcmc.R	99.08% <100%> (+0.1%)	⬆️
R/tidy-params.R	100% <100%> (ø)
R/ppc-test-statistics.R	100% <0%> (ø)	⬆️
R/mcmc-traces.R	100% <0%> (ø)	⬆️
R/ppc-loo.R	100% <0%> (ø)	⬆️
R/mcmc-distributions.R	100% <0%> (ø)	⬆️
R/mcmc-recover.R	100% <0%> (ø)	⬆️
... and 3 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a947141...a1c1eec. Read the comment docs.

[jgabry]

Given that this PR makes it possible to use the tidyselect helpers (e.g., starts_with(), num_range(), etc.) to select parameters, does it make sense to provide some additional helpers that are convenient?

For example, if parameters are named "beta1", "beta2", ..., "beta10" and we want to select just 2 through 7 then the num_range() tidyselect helper works great:

mcmc_hist(..., pars = vars(num_range("beta", 2:7)))

but num_range() doesn't work if the parameters have brackets in their names like "beta[1]", "beta[2]", ..., "beta[10]", which is far more common. But we could conceivably export a helper function like this:

param_range <- function(prefix, range) {
  nms <- paste0(prefix, "[", range, "]")
  param_matches <- match(nms, table = tidyselect::peek_vars())
  param_matches[!is.na(param_matches)]
}

which would enable

mcmc_hist(..., pars = vars(param_range("beta", 2:7)))

[tjmahr]

num_range() doesn't work if the parameters have brackets in their names like "beta[1]", "beta[2]", ..., "beta[10]", which is far more common. But we could conceivably export a helper function like this:

This is great.

[jgabry]

Ok cool I’ll add that to the PR.

[tjmahr]

I was worried about mixed models where you get a lot going on the brackets, so I tried to make a general one. This would require glue, but glue is dependency-less and it is required by most of the tidyverse packages that we already depend on.

library(tidyverse)
#> Registered S3 methods overwritten by 'ggplot2':
#>   method         from 
#>   [.quosures     rlang
#>   c.quosures     rlang
#>   print.quosures rlang
d <- structure(list(b_Intercept = c(0.25, 0.13, 0.77, 0.67, 0.44, 
0.33), sd_condition__Intercept = c(1.01, 1.01, 0.78, 0.77, 0.89, 
0.76), sigma = c(0.63, 0.54, 0.62, 0.59, 0.47, 0.44), `r_condition[A,Intercept]` = c(-0.2, 
0.16, -0.82, -0.28, -0.53, -0.25), `r_condition[B,Intercept]` = c(0.98, 
0.53, 0.44, 0.15, 0.74, 0.41), `r_condition[C,Intercept]` = c(1.33, 
1.79, 0.83, 1.5, 1.15, 1.54), `r_condition[A,Slope]` = c(0.86, 
0.78, 0.19, -0.14, 0.87, 0.49), `r_condition[B,Slope]` = c(-1.18, 
-0.96, -1.66, -1.62, -1.53, -1.1), lp__ = c(-51.66, -51.15, -53.32, 
-56.79, -56.48, -54.96)), class = c("tbl_df", "tbl", "data.frame"
), row.names = c(NA, -6L))
  
param_glue <- function(pattern, ...) {
  dots <- as.list(expand.grid(...))
  nms <- as.character(glue::glue_data(dots, pattern))

  param_matches <- match(nms, tidyselect::peek_vars())
  param_matches[!is.na(param_matches)]
}

d
#> # A tibble: 6 x 9
#>   b_Intercept sd_condition__I~ sigma `r_condition[A,~ `r_condition[B,~
#>         <dbl>            <dbl> <dbl>            <dbl>            <dbl>
#> 1        0.25             1.01  0.63            -0.2              0.98
#> 2        0.13             1.01  0.54             0.16             0.53
#> 3        0.77             0.78  0.62            -0.82             0.44
#> 4        0.67             0.77  0.59            -0.28             0.15
#> 5        0.44             0.89  0.47            -0.53             0.74
#> 6        0.33             0.76  0.44            -0.25             0.41
#> # ... with 4 more variables: `r_condition[C,Intercept]` <dbl>,
#> #   `r_condition[A,Slope]` <dbl>, `r_condition[B,Slope]` <dbl>, lp__ <dbl>

d %>% 
  select(
    param_glue(
      "r_condition[{level},Intercept]", 
      level = c("A", "B"))
  )
#> # A tibble: 6 x 2
#>   `r_condition[A,Intercept]` `r_condition[B,Intercept]`
#>                        <dbl>                      <dbl>
#> 1                      -0.2                        0.98
#> 2                       0.16                       0.53
#> 3                      -0.82                       0.44
#> 4                      -0.28                       0.15
#> 5                      -0.53                       0.74
#> 6                      -0.25                       0.41

d %>% 
  select(
    param_glue(
      "r_condition[{level},{type}]", 
      level = c("A", "B"), 
      type = c("Intercept", "Slope"))
    )
#> # A tibble: 6 x 4
#>   `r_condition[A,In~ `r_condition[B,In~ `r_condition[A,S~ `r_condition[B,S~
#>                <dbl>              <dbl>             <dbl>             <dbl>
#> 1              -0.2                0.98              0.86             -1.18
#> 2               0.16               0.53              0.78             -0.96
#> 3              -0.82               0.44              0.19             -1.66
#> 4              -0.28               0.15             -0.14             -1.62
#> 5              -0.53               0.74              0.87             -1.53
#> 6              -0.25               0.41              0.49             -1.1

^{Created on 2019-05-20 by the reprex package (v0.3.0)}

[jgabry]

Good idea!

[jgabry]

I’ll incorporate that.

[jgabry]

I think this is ready to go.

[tjmahr]

I expanded the documentation to tidy-params to include all the select() gotchas I could think of. Everything else seems fine.

[jgabry]

Awesome, thanks!