fix geom_smooth() and add regression tests. Closes #2621. (#2626)

* fix geom_smooth() and add regression tests. Closes #2621.

* tidyverse style

* fix news

Author: clauswilke

NEWS.md

@@ -343,10 +343,8 @@
   plotting thick arrows (@Ax3man, #774).
 
 * `geom_smooth()` now reports the formula used when `method = "auto"` 
-  (@davharris #1951). When mgcv is used, the `s` in `y ~ s(x, bs = "cs")` 
-  is now fully qualified as `mgcv::s()` to avoid name conflicts (@bfgray3, 
-  #2535). `geom_smooth()` now orders by the `x` aesthetic, making it easier 
-  to pass pre-computed values without manual ordering (@izahn, #2028). It 
+  (@davharris #1951). `geom_smooth()` now orders by the `x` aesthetic, making it 
+  easier to pass pre-computed values without manual ordering (@izahn, #2028). It 
   also now knows it has `ymin` and `ymax` aesthetics (#1939). The legend 
   correctly reflects the status of the `se` argument when used with stats 
   other than the default (@clauswilke, #1546).

R/stat-smooth.r

@@ -83,9 +83,9 @@ StatSmooth <- ggproto("StatSmooth", Stat,
         params$method <- "loess"
       } else {
         params$method <- "gam"
-        params$formula <- y ~ mgcv::s(x, bs = "cs")
+        params$formula <- y ~ s(x, bs = "cs")
       }
-      message("`geom_smooth()` using method = '", params$method, 
+      message("`geom_smooth()` using method = '", params$method,
               "' and formula '", deparse(params$formula), "'")
     }
     if (identical(params$method, "gam")) {

tests/testthat/test-geom-smooth.R

@@ -9,6 +9,48 @@ test_that("Data is ordered by x", {
   expect_equal(layer_data(ps)[c("x", "y")], df[order(df$x), ])
 })
 
+test_that("Default smoothing methods for small and large data sets work", {
+  # test small data set
+  set.seed(6531)
+  x <- rnorm(10)
+  df <- data.frame(
+    x = x,
+    y = x^2 + 0.5 * rnorm(10)
+  )
+
+  m <- loess(y ~ x, data = df, span = 0.75)
+  range <- range(df$x, na.rm = TRUE)
+  xseq <- seq(range[1], range[2], length.out = 80)
+  out <- predict(m, data.frame(x = xseq))
+  p <- ggplot(df, aes(x, y)) + geom_smooth()
+
+  expect_message(
+    plot_data <- layer_data(p),
+    "method = 'loess' and formula 'y ~ x'"
+  )
+  expect_equal(plot_data$y, as.numeric(out))
+
+  # test large data set
+  x <- rnorm(1001) # 1000 is the cutoff point for gam
+  df <- data.frame(
+    x = x,
+    y = x^2 + 0.5 * rnorm(1001)
+  )
+
+  m <- mgcv::gam(y ~ s(x, bs = "cs"), data = df)
+  range <- range(df$x, na.rm = TRUE)
+  xseq <- seq(range[1], range[2], length.out = 80)
+  out <- predict(m, data.frame(x = xseq))
+  p <- ggplot(df, aes(x, y)) + geom_smooth()
+
+  expect_message(
+    plot_data <- layer_data(p),
+    "method = 'gam' and formula 'y ~ s\\(x, bs = \"cs\"\\)"
+  )
+  expect_equal(plot_data$y, as.numeric(out))
+
+})
+
 # Visual tests ------------------------------------------------------------
 
 test_that("geom_smooth() works with alternative stats", {