Merge pull request #328 from ishaan-arora-1/fix/clarify-print-dims-198

Clarify print output to specify posterior draws and log-likelihood terms

Author: florence-bockting

NEWS.md

@@ -1,5 +1,8 @@
 # loo (development version)
 
+* Update user messages in `print()` by @ishaan-arora-1, @florence-bockting in 
+#328.
+
 # loo 2.10.1
 
 * Revert behavior of internal `psis_smooth_tail()` function that led to several

R/elpd.R

@@ -67,9 +67,9 @@ elpd_object <- function(pointwise, dims) {
 }
 #' @export
 print_dims.elpd_generic <- function(x, ...) {
+  dims <- dim(x)
   cat(
-    "Computed from",
-    paste(dim(x), collapse = " by "),
-    "log-likelihood matrix using the generic elpd function\n"
+    "Computed from", dims[1], "posterior draws and",
+    dims[2], "log-likelihood terms using the generic elpd function.\n"
   )
 }

R/print.R

@@ -141,40 +141,40 @@ print_dims <- function(x, ...) UseMethod("print_dims")
 #' @rdname print_dims
 #' @export
 print_dims.importance_sampling <- function(x, ...) {
+  dims <- dim(x)
   cat(
-    "Computed from",
-    paste(dim(x), collapse = " by "),
-    "log-weights matrix.\n"
+    "Computed from", dims[1], "posterior draws and",
+    dims[2], "log-weight terms.\n"
   )
 }
 
 #' @rdname print_dims
 #' @export
 print_dims.psis_loo <- function(x, ...) {
+  dims <- dim(x)
   cat(
-    "Computed from",
-    paste(dim(x), collapse = " by "),
-    "log-likelihood matrix.\n"
+    "Computed from", dims[1], "posterior draws and",
+    dims[2], "log-likelihood terms.\n"
   )
 }
 
 #' @rdname print_dims
 #' @export
 print_dims.importance_sampling_loo <- function(x, ...) {
+  dims <- dim(x)
   cat(
-    "Computed from",
-    paste(dim(x), collapse = " by "),
-    "log-likelihood matrix using", class(x)[1], ".\n"
+    "Computed from", dims[1], "posterior draws and",
+    dims[2], "log-likelihood terms using", class(x)[1], ".\n"
   )
 }
 
 #' @rdname print_dims
 #' @export
 print_dims.waic <- function(x, ...) {
+  dims <- dim(x)
   cat(
-    "Computed from",
-    paste(dim(x), collapse = " by "),
-    "log-likelihood matrix.\n"
+    "Computed from", dims[1], "posterior draws and",
+    dims[2], "log-likelihood terms.\n"
   )
 }
 
@@ -191,11 +191,13 @@ print_dims.kfold <- function(x, ...) {
 #' @export
 print_dims.psis_loo_ss <- function(x, ...) {
   cat(
-    "Computed from",
-    paste(c(dim(x)[1], nobs(x)) , collapse = " by "),
-    "subsampled log-likelihood\nvalues from",
+    "Subsampling loo uses 1) faster, more biased computation with ",
     length(x$loo_subsampling$elpd_loo_approx),
-    "total observations.\n"
+    " log-lik.\n",
+    "terms, and 2) slower, more accurate computation with ",
+    nobs(x),
+    " subsampled terms.\n",
+    sep = ""
   )
 }
 

tests/testthat/_snaps/loo_moment_matching.md

@@ -13,7 +13,7 @@
       Some Pareto k diagnostic values are too high. See help('pareto-k-diagnostic') for details.
     Output
       
-      Computed from 4000 by 30 log-likelihood matrix.
+      Computed from 4000 posterior draws and 30 log-likelihood terms.
       
                Estimate   SE
       elpd_loo    -74.0 18.8

tests/testthat/_snaps/loo_subsampling_cases.md

@@ -4,8 +4,8 @@
       print(looss_1)
     Output
       
-      Computed from 4000 by 100 subsampled log-likelihood
-      values from 3020 total observations.
+      Subsampling loo uses 1) faster, more biased computation with 3020 log-lik.
+      terms, and 2) slower, more accurate computation with 100 subsampled terms.
       
                Estimate   SE subsampling SE
       elpd_loo  -1968.5 15.6            0.3
@@ -24,8 +24,8 @@
       print(looss_1b)
     Output
       
-      Computed from 4000 by 200 subsampled log-likelihood
-      values from 3020 total observations.
+      Subsampling loo uses 1) faster, more biased computation with 3020 log-lik.
+      terms, and 2) slower, more accurate computation with 200 subsampled terms.
       
                Estimate   SE subsampling SE
       elpd_loo  -1968.3 15.6            0.2
@@ -44,7 +44,7 @@
       print(aploo_1)
     Output
       
-      Computed from 2000 by 3020 log-likelihood matrix.
+      Computed from 2000 posterior draws and 3020 log-likelihood terms.
       
                Estimate   SE
       elpd_loo  -1968.4 15.6
@@ -64,8 +64,8 @@
       print(looapss_1)
     Output
       
-      Computed from 2000 by 100 subsampled log-likelihood
-      values from 3020 total observations.
+      Subsampling loo uses 1) faster, more biased computation with 3020 log-lik.
+      terms, and 2) slower, more accurate computation with 100 subsampled terms.
       
                Estimate   SE subsampling SE
       elpd_loo  -1968.2 15.6            0.4
@@ -85,8 +85,8 @@
       print(looss_2)
     Output
       
-      Computed from 4000 by 100 subsampled log-likelihood
-      values from 3020 total observations.
+      Subsampling loo uses 1) faster, more biased computation with 3020 log-lik.
+      terms, and 2) slower, more accurate computation with 100 subsampled terms.
       
                Estimate   SE subsampling SE
       elpd_loo  -1952.0 16.2            0.2

tests/testthat/_snaps/print_plot.md

@@ -4,7 +4,7 @@
       print(waic1)
     Output
       
-      Computed from 1000 by 32 log-likelihood matrix.
+      Computed from 1000 posterior draws and 32 log-likelihood terms.
       
                 Estimate  SE
       elpd_waic    -83.5 4.3
@@ -18,7 +18,7 @@
     Code
       print(psis1)
     Output
-      Computed from 1000 by 32 log-weights matrix.
+      Computed from 1000 posterior draws and 32 log-weight terms.
       MCSE and ESS estimates assume independent draws (r_eff=1).
       
       All Pareto k estimates are good (k < 0.67).
@@ -30,7 +30,7 @@
       print(loo1)
     Output
       
-      Computed from 1000 by 32 log-likelihood matrix.
+      Computed from 1000 posterior draws and 32 log-likelihood terms.
       
                Estimate  SE
       elpd_loo    -83.6 4.3
@@ -49,7 +49,7 @@
       print(loo1_r_eff)
     Output
       
-      Computed from 1000 by 32 log-likelihood matrix.
+      Computed from 1000 posterior draws and 32 log-likelihood terms.
       
                Estimate  SE
       elpd_loo    -83.6 4.3

tests/testthat/_snaps/psis.md

@@ -4787,7 +4787,7 @@
     Message
       Replacing NAs in `r_eff` with 1s
     Output
-      Computed from 1000 by 32 log-weights matrix.
+      Computed from 1000 posterior draws and 32 log-weight terms.
       MCSE and ESS estimates assume MCMC draws (r_eff in [0.6, 1.0]).
       
       All Pareto k estimates are good (k < 0.67).
@@ -4803,7 +4803,7 @@
       Warning:
       Some Pareto k diagnostic values are too high. See help('pareto-k-diagnostic') for details.
     Output
-      Computed from 10 by 32 log-weights matrix.
+      Computed from 10 posterior draws and 32 log-weight terms.
       MCSE and ESS estimates assume independent draws (r_eff=1).
       Pareto k diagnostic values:
                              Count Pct.    Min. ESS

tests/testthat/_snaps/tisis.md

@@ -5,7 +5,7 @@
     Message
       Replacing NAs in `r_eff` with 1s
     Output
-      Computed from 1000 by 32 log-weights matrix.
+      Computed from 1000 posterior draws and 32 log-weight terms.
       MCSE and ESS estimates assume MCMC draws (r_eff in [0.6, 1.0]).
       
       All Pareto k estimates are good (k < 0.67).
@@ -17,7 +17,7 @@
       print(loo_tis)
     Output
       
-      Computed from 1000 by 32 log-likelihood matrix using tis_loo .
+      Computed from 1000 posterior draws and 32 log-likelihood terms using tis_loo .
       
                Estimate  SE
       elpd_loo    -83.6 4.3
@@ -31,7 +31,7 @@
       print(loo_sis)
     Output
       
-      Computed from 1000 by 32 log-likelihood matrix using sis_loo .
+      Computed from 1000 posterior draws and 32 log-likelihood terms using sis_loo .
       
                Estimate  SE
       elpd_loo    -83.6 4.3

tests/testthat/test_loo_subsampling_cases.R

@@ -171,25 +171,25 @@ test_that("Test loo_subsampling and loo_approx with radon data", {
   ))
   expect_failure(expect_output(
     print(full_loo),
-    "subsampled log-likelihood\nvalues"
+    "subsampled log-likelihood terms"
   ))
 
   expect_failure(expect_output(
     print(loo_ss),
     "Posterior approximation correction used\\."
   ))
-  expect_output(print(loo_ss), "subsampled log-likelihood\nvalues")
+  expect_output(print(loo_ss), "Subsampling loo uses 1) faster,")
 
   expect_output(print(loo_ap_ss), "Posterior approximation correction used\\.")
-  expect_output(print(loo_ap_ss), "subsampled log-likelihood\nvalues")
+  expect_output(print(loo_ap_ss), "Subsampling loo uses 1) faster,")
 
   expect_output(
     print(loo_ap_ss_full),
     "Posterior approximation correction used\\."
   )
   expect_failure(expect_output(
     print(loo_ap_ss_full),
-    "subsampled log-likelihood\nvalues"
+    "subsampled log-likelihood terms"
   ))
 
   # Test conversion of objects
@@ -369,11 +369,8 @@ test_that("Test the vignette", {
       loo_approximation_draws = 100
     )
   )
-  expect_output(
-    print(looss_2),
-    "Computed from 4000 by 100 subsampled log-likelihood"
-  )
-  expect_output(print(looss_2), "values from 3020 total observations.")
+  expect_output(print(looss_2), "Subsampling loo uses 1) faster,")
+  expect_output(print(looss_2), "Subsampling loo uses 1) faster,")
   expect_output(
     print(looss_2),
     "MCSE and ESS estimates assume independent draws"

vignettes/loo2-large-data.Rmd

@@ -188,8 +188,8 @@ print(loo_ss_1)
 ```
 
 ```
-Computed from 4000 by 100 subsampled log-likelihood
-values from 3020 total observations.
+Subsampling loo uses 1) faster, more biased computation with 3020 log-lik.
+terms, and 2) slower, more accurate computation with 100 subsampled terms.
 
          Estimate   SE subsampling SE
 elpd_loo  -1968.5 15.6            0.3
@@ -235,13 +235,13 @@ loo_ss_1b <-
     r_eff = r_eff,
     draws = parameter_draws_1,
     data = stan_df_1
-  ) 
+  )
 print(loo_ss_1b)
 ```
 
 ```
-Computed from 4000 by 200 subsampled log-likelihood
-values from 3020 total observations.
+Subsampling loo uses 1) faster, more biased computation with 3020 log-lik.
+terms, and 2) slower, more accurate computation with 200 subsampled terms.
 
          Estimate   SE subsampling SE
 elpd_loo  -1968.3 15.6            0.2
@@ -285,8 +285,8 @@ print(loo_ss_1c)
 
 
 ```
-Computed from 4000 by 100 subsampled log-likelihood
-values from 3020 total observations.
+Subsampling loo uses 1) faster, more biased computation with 3020 log-lik.
+terms, and 2) slower, more accurate computation with 100 subsampled terms.
 
          Estimate   SE subsampling SE
 elpd_loo  -1968.9 15.4            0.5
@@ -339,7 +339,7 @@ print(loo_ap_1)
 The function creates a class, `psis_loo_ap` that inherits from `psis_loo, loo`.
 
 ```
-Computed from 2000 by 3020 log-likelihood matrix
+Computed from 2000 posterior draws and 3020 log-likelihood terms
 
          Estimate   SE
 elpd_loo  -1968.4 15.6
@@ -375,8 +375,8 @@ print(loo_ap_ss_1)
 ```
 
 ```
-Computed from 2000 by 100 subsampled log-likelihood
-values from 3020 total observations.
+Subsampling loo uses 1) faster, more biased computation with 3020 log-lik.
+terms, and 2) slower, more accurate computation with 100 subsampled terms.
 
          Estimate   SE subsampling SE
 elpd_loo  -1968.2 15.6            0.4
@@ -455,8 +455,8 @@ print(loo_ss_2)
 ```
 
 ```
-Computed from 4000 by 100 subsampled log-likelihood
-values from 3020 total observations.
+Subsampling loo uses 1) faster, more biased computation with 3020 log-lik.
+terms, and 2) slower, more accurate computation with 200 subsampled terms.
 
          Estimate   SE subsampling SE
 elpd_loo  -1952.0 16.2            0.2