fix: flatline_forecaster duplicate predictions on trailing-NA input (#454)

DESCRIPTION

@@ -1,6 +1,6 @@
 Package: epipredict
 Title: Basic epidemiology forecasting methods
-Version: 0.2.3
+Version: 0.2.4
 Authors@R: c(
     person("Daniel J.", "McDonald", , "daniel@stat.ubc.ca", role = c("aut", "cre")),
     person("Ryan", "Tibshirani", , "ryantibs@cmu.edu", role = "aut"),

NEWS.md

@@ -2,6 +2,10 @@
 
 Pre-1.0.0 numbering scheme: 0.x will indicate releases, while 0.0.x will indicate PR's.
 
+# epipredict 0.2.4
+
+- Fix `flatline_forecaster()` to return one prediction per geographic key when the input `epi_df` has trailing rows with `NA`s in the outcome (#454). Previously, the forecast was duplicated once per trailing-NA day.
+
 # epipredict 0.2.3
 
 - Fix `print.canned_epipred()` so the latency-adjustment information actually displays for canned forecasters that include `step_adjust_latency` in their recipe (#447).

R/get_test_data.R

@@ -66,13 +66,13 @@ get_test_data <- function(recipe, x) {
 
   # If we skip NA completion, we remove undesirably early time values
   # Happens globally, over all groups
-  x <- filter(x, max_time_value - time_value <= keep)
+  x <- filter(x, max_time_value - time_value <= keep, time_value <= max_time_value)
 
   # If all(lags > 0), then we get rid of recent data
   if (min_lags > 0 && min_lags < Inf) {
     x <- filter(x, max_time_value - time_value >= min_lags)
   }
 
-  filter(x, max_time_value - time_value <= keep) %>%
+  filter(x, max_time_value - time_value <= keep, time_value <= max_time_value) %>%
     epiprocess::ungroup()
 }

tests/testthat/test-flatline_forecaster.R

@@ -0,0 +1,22 @@
+test_that("flatline_forecaster returns one prediction per geo with trailing NAs (issue #454)", {
+  jhu <- epidatasets::covid_case_death_rates %>%
+    dplyr::filter(time_value >= as.Date("2021-11-01"))
+  geos <- unique(jhu$geo_value)
+  one_day <- tibble::tibble(
+    geo_value = geos,
+    time_value = as.Date("2022-01-01"),
+    case_rate = NA_real_,
+    death_rate = 0
+  )
+  two_day <- one_day %>% dplyr::mutate(time_value = as.Date("2022-01-02"))
+  jhu_nad <- jhu %>%
+    tibble::as_tibble() %>%
+    dplyr::bind_rows(one_day, two_day) %>%
+    epiprocess::as_epi_df()
+
+  res <- flatline_forecaster(jhu_nad, "case_rate")
+
+  expect_equal(nrow(res$predictions), length(geos))
+  counts <- res$predictions %>% dplyr::count(geo_value, target_date)
+  expect_true(all(counts$n == 1L))
+})