Skip to contents

Run nowcasts for multiple states/locations

Source: R/run_nowcasts.R
run_state_nowcasts.Rd

Orchestrates nowcast generation for multiple geographic locations, with optional parallel processing. Wraps each location in error handling to continue processing even if individual locations fail.

This function handles the full pipeline from cumulative reporting data (as returned by fetch_reporting_data()) to nowcast results, including conversion to incremental counts internally.

Usage

run_state_nowcasts(
  data,
  config,
  location_configs = NULL,
  locations = "all",
  nowcast_date = NULL,
  cumulative = TRUE,
  parallel = FALSE
)

Arguments

data

A data frame of reporting data as returned by fetch_reporting_data(), containing columns reference_date, report_date, count, location, and signal.

config

A nowcast_config object to use as the default for all locations.

location_configs

Optional named list of nowcast_config objects keyed by location code (e.g., list(ca = config_ca, ny = config_ny)). These override the default config for specific locations.

locations

Character vector. Either "all" for all available locations in the data, or a character vector of specific location codes (e.g., c("ca", "ny")). Default is "all".

nowcast_date

Date. The date to use as the nowcast date. Defaults to the maximum report_date in the data.

cumulative

Logical. If TRUE (the default), the input data is assumed to be cumulative counts (as returned by fetch_reporting_data()) and will be converted to incremental counts using cumulative_to_incremental(). Set to FALSE if the data is already incremental.

parallel

Logical. Whether to use parallel processing via purrr::in_parallel(). Requires the mirai package. Default is FALSE.

Value

A data frame with columns:

  • reference_date: The reference date being nowcasted

  • location: Geographic identifier

  • pred_count: Predicted count (one row per draw)

  • draw: Draw number

  • output_type: Type of output ("samples" or "point")

The returned data frame has an attribute failed_locations containing a character vector of any locations that failed to process.

Examples

if (FALSE) { # \dontrun{
# Fetch cumulative data for states
source <- delphi_epidata_source(target = "covid", geo_types = "state")
data <- fetch_reporting_data(source, reference_dates, report_dates)

# Single config for all locations
config <- nowcast_config()
results <- run_state_nowcasts(data, config, locations = c("ca", "ny"))

# Override config for specific locations
config <- nowcast_config(draws = 500)
location_configs <- list(
  ca = nowcast_config(max_delay = 5, draws = 500),
  ny = nowcast_config(max_delay = 7, uncertainty_model = "skellam", draws = 500)
)
results <- run_state_nowcasts(
  data, config,
  location_configs = location_configs,
  locations = c("ca", "ny", "tx")
)
# ca and ny use their specific configs, tx uses the default

# Check for failures
attr(results, "failed_locations")
} # }