Skip to contents

Convert EpiNow2 model output to a forecast_sample object

Source: R/as_forecast_sample.R
as_forecast_sample.Rd

[Experimental] Convert outputs of EpiNow2 fitting and forecasting functions to forecast_sample objects via scoringutils::as_forecast_sample() for evaluating predictive performance. Methods are provided for objects returned by epinow(), estimate_infections(), forecast_secondary(), and estimate_truncation().

These methods extract sample-level posterior predictions via get_predictions() with format = "sample", merge them with the supplied observations on date, and pass the result to scoringutils::as_forecast_sample().

scoringutils is an optional dependency; calling these methods without it installed gives an informative error.

Usage

# S3 method for class 'estimate_infections'
as_forecast_sample(data, observations, horizon = 0, ...)

# S3 method for class 'epinow'
as_forecast_sample(data, observations, horizon = 0, ...)

# S3 method for class 'forecast_secondary'
as_forecast_sample(data, observations, horizon = 0, ...)

# S3 method for class 'estimate_truncation'
as_forecast_sample(data, observations, horizon = -Inf, ...)

Arguments

data

Output of epinow(), estimate_infections(), forecast_secondary(), or estimate_truncation().

observations

A <data.frame> of observed values to score against. Must contain a date column. For epinow() and estimate_infections() objects must also contain a confirm column; for forecast_secondary() objects a secondary column; for estimate_truncation() objects a confirm column representing the latest, least-truncated observations.

horizon

Numeric scalar lower bound on the horizon column of get_predictions() output. Predictions with a horizon value at or above this bound are retained. Defaults to 0 for epinow(), estimate_infections() and forecast_secondary() (i.e. forecast period only) and to -Inf for estimate_truncation() (keep all reconstructed horizons). Pass horizon = -Inf to disable filtering.

...

Additional arguments passed to scoringutils::as_forecast_sample(). forecast_unit is set automatically from the object class (forecast_date, date, horizon, plus dataset for estimate_truncation()) and cannot be overridden.

Value

A forecast_sample object as returned by scoringutils::as_forecast_sample(). Rows for which observations does not provide a value on the corresponding date are dropped.

See also

get_predictions() for the underlying sample extraction.

Examples

# \donttest{
library(scoringutils)

# samples and calculation time have been reduced for this example
# for real analyses, use at least samples = 2000
fit <- estimate_infections(example_confirmed[1:40],
  generation_time = gt_opts(example_generation_time),
  delays = delay_opts(example_incubation_period + example_reporting_delay),
  rt = rt_opts(prior = LogNormal(mean = 2, sd = 0.2)),
  stan = stan_opts(samples = 100, warmup = 200)
)
#> Warning: The largest R-hat is NA, indicating chains have not mixed.
#> Running the chains for more iterations may help. See
#> https://mc-stan.org/misc/warnings.html#r-hat
#> Warning: Bulk Effective Samples Size (ESS) is too low, indicating posterior means and medians may be unreliable.
#> Running the chains for more iterations may help. See
#> https://mc-stan.org/misc/warnings.html#bulk-ess
#> Warning: Tail Effective Samples Size (ESS) is too low, indicating posterior variances and tail quantiles may be unreliable.
#> Running the chains for more iterations may help. See
#> https://mc-stan.org/misc/warnings.html#tail-ess

forecast_obj <- as_forecast_sample(fit, observations = example_confirmed)
score(forecast_obj)
#> Warning: Predictions appear to be integer-valued.
#> ! The log score uses kernel density estimation, which may not be appropriate
#>   for integer-valued forecasts.
#> ℹ See the scoringRules package for alternatives for discrete probability
#>   distributions.
#>    forecast_date       date horizon  bias      dss     crps overprediction
#>           <Date>     <Date>   <num> <num>    <num>    <num>          <num>
#> 1:    2020-04-01 2020-04-01       0 -0.12 14.02019 248.8000           0.00
#> 2:    2020-04-01 2020-04-02       1 -0.28 14.49945 376.5825           0.00
#> 3:    2020-04-01 2020-04-03       2  0.12 14.76330 346.7009          11.58
#> 4:    2020-04-01 2020-04-04       3 -0.26 14.63412 383.9739           0.00
#> 5:    2020-04-01 2020-04-05       4 -0.16 14.82256 356.3254           0.00
#> 6:    2020-04-01 2020-04-06       5 -0.06 14.79280 370.4198           0.00
#> 7:    2020-04-01 2020-04-07       6  0.02 14.61886 330.1965           0.22
#> 8:    2020-04-01 2020-04-08       7 -0.14 14.18907 293.5545           0.00
#>    underprediction dispersion log_score      mad ae_median     se_mean
#>              <num>      <num>     <num>    <num>     <num>       <num>
#> 1:           10.12   238.6800  7.955772 1097.865     167.0   2242.9696
#> 2:           74.88   301.7025  8.273460 1220.180     473.5 105410.6089
#> 3:            0.00   335.1209  8.208060 1441.828     148.0 204982.5625
#> 4:           73.24   310.7339  8.252185 1245.384     559.5 145702.5241
#> 5:           33.32   323.0054  8.181045 1221.662     345.5   1740.5584
#> 6:            1.70   368.7198  8.356224 1604.914      73.5  28083.0564
#> 7:            0.00   329.9765  8.246860 1343.236      17.5  49858.4241
#> 8:           13.78   279.7745  8.145034 1192.752     210.5    588.0625
# }