layer_add_target_date: Postprocessing step to add the target date

In cmu-delphi/epipredict: Basic epidemiology forecasting methods

Postprocessing step to add the target date

Description

Postprocessing step to add the target date

Usage

layer_add_target_date(
  frosting,
  target_date = NULL,
  id = rand_id("add_target_date")
)

Arguments

frosting	a frosting postprocessor
target_date	The target date to add as a column to the epi_df. If there's a forecast date specified upstream (either in a step_adjust_latency or in a layer_forecast_date), then it is the forecast date plus ahead (from step_epi_ahead in the epi_recipe). Otherwise, it is the maximum time_value (from the data used in pre-processing, fitting the model, and postprocessing) plus ahead, where ahead has been specified in preprocessing. The user may override these by specifying a target date of their own (of the form "yyyy-mm-dd").
id	a random id string

Details

By default, this function assumes that a value for ahead has been specified in a preprocessing step (most likely in step_epi_ahead). Then, ahead is added to the forecast_date in the test data to get the target date. forecast_date can be set in 3 ways:

1. step_adjust_latency, which typically uses the training epi_df's as_of

2. layer_add_forecast_date, which inherits from 1 if not manually specifed

3. if none of those are the case, it is simply the maximum time_value over every dataset used (prep, training, and prediction).

Value

an updated frosting postprocessor

Examples

jhu <- covid_case_death_rates %>%
  filter(time_value > "2021-11-01", geo_value %in% c("ak", "ca", "ny"))
r <- epi_recipe(jhu) %>%
  step_epi_lag(death_rate, lag = c(0, 7, 14)) %>%
  step_epi_ahead(death_rate, ahead = 7) %>%
  step_epi_naomit()

wf <- epi_workflow(r, linear_reg()) %>% fit(jhu)

# Use ahead + forecast date
f <- frosting() %>%
  layer_predict() %>%
  layer_add_forecast_date(forecast_date = as.Date("2022-05-31")) %>%
  layer_add_target_date() %>%
  layer_naomit(.pred)
wf1 <- wf %>% add_frosting(f)

p <- forecast(wf1)
p

# Use ahead + forecast_date from adjust_latency
# setting the `as_of` to something realistic
attributes(jhu)$metadata$as_of <- max(jhu$time_value) + 3
r <- epi_recipe(jhu) %>%
  step_epi_lag(death_rate, lag = c(0, 7, 14)) %>%
  step_epi_ahead(death_rate, ahead = 7) %>%
  step_adjust_latency(method = "extend_ahead") %>%
  step_epi_naomit()
f2 <- frosting() %>%
  layer_predict() %>%
  layer_add_target_date() %>%
  layer_naomit(.pred)
wf2 <- wf %>% add_frosting(f2)

p2 <- forecast(wf2)
p2

# Use ahead + max time value from pre, fit, post
# which is the same if include `layer_add_forecast_date()`
f3 <- frosting() %>%
  layer_predict() %>%
  layer_add_target_date() %>%
  layer_naomit(.pred)
wf3 <- wf %>% add_frosting(f3)

p3 <- forecast(wf2)
p2

# Specify own target date
f4 <- frosting() %>%
  layer_predict() %>%
  layer_add_target_date(target_date = "2022-01-08") %>%
  layer_naomit(.pred)
wf4 <- wf %>% add_frosting(f4)

p4 <- forecast(wf4)
p4

cmu-delphi/epipredict documentation built on March 5, 2025, 12:17 p.m.