MEM intensity thresholds
Usage
mem_thresholds_v1(x, ...)
# S3 method for class 'csfmt_ensemble_v3'
mem_thresholds_v1(
x,
measure,
min_seasons = 2,
prefer_seasons = 5,
i.seasons = 10,
min_weeks_per_season = 30,
exclude_seasons = NULL,
...
)Arguments
- x
Data object.
- ...
Passed to methods.
- measure
The `$draws` measure to threshold on (a rate or count).
- min_seasons
Hard floor of complete prior seasons needed to fit.
- prefer_seasons
Preferred training depth (provisional below this).
- i.seasons
Max seasons passed to mem::memmodel.
- min_weeks_per_season
Weeks needed for a season to count as training.
- exclude_seasons
Optional character vector of seasons (e.g. `c("2009/2010", "2019/2020")`, the `isoyearweek_to_season_c` form) to drop from the MEM training baseline – anomalous seasons (pandemic years, data gaps) that would distort the thresholds. Thresholds are still ESTIMATED for every season (including excluded ones) from its remaining non-excluded prior seasons; only the baseline they are fit on changes.
Value
The `csfmt_ensemble_v3` with per-draw MEM intensity columns added to `$draws`. The added columns are the ordinal 1..5 status for `measure` and its threshold levels, so the intensity level propagates through the later quantile collapse.
See also
vignette("pipeline", package = "csalert"), which runs this
function as stage 6 of its pipeline, on a five-season synthetic series.
Examples
# MEM needs several complete prior seasons, so this fixture spans four:
# 212 weeks with a winter peak in each.
if (requireNamespace("mem", quietly = TRUE)) {
w <- cstime::dates_by_isoyearweek$isoyearweek
i <- match("2018-30", w)
n <- 212L
iyw <- w[i + 0:(n - 1)]
set.seed(1)
lam <- 20 + 60 * exp(-(cstime::isoyearweek_to_seasonweek_n(iyw) - 25)^2 / 40)
ens <- csfmt_ensemble_v3(
data.table::data.table(
isoyearweek = iyw, location_code = "nation", age = "total"
),
id_cols = c("location_code", "age"),
draws = list(numerator_nowcasted = matrix(rpois(n * 50, lam), nrow = n))
)
ens <- mem_thresholds_v1(ens, measure = "numerator_nowcasted")
# thresholds are estimated leave-future-out, so the early seasons get none
print(unique(ens$data[
!is.na(mem_high),
.(
season = cstime::isoyearweek_to_season_c(isoyearweek),
mem_preepidemic, mem_medium, mem_high, mem_veryhigh, mem_n_seasons
)
]))
# A week whose season has no thresholds gets NA for every draw, so it is not
# classified at all. On this fixture that is the first two seasons:
status <- ens$draws$numerator_nowcasted_status
print(c(
weeks = nrow(status),
weeks_with_no_threshold = sum(apply(status, 1, function(r) all(is.na(r))))
))
# For a week that DOES have thresholds, every draw is classified, so the
# alert level arrives as a distribution rather than as a single label.
r <- ens_collapse(ens, probs = 0.5)
pcols <- grep("_status_prob_", names(r), value = TRUE)
print(r[
isoyearweek %in% c("2021-45", "2022-02"),
c("isoyearweek", pcols),
with = FALSE
])
}
#> mem_thresholds_v1: 3 season(s) fit on < 5 training seasons (provisional); see mem_n_seasons.
#> season mem_preepidemic mem_medium mem_high mem_veryhigh mem_n_seasons
#> <char> <num> <num> <num> <num> <int>
#> 1: 2020/2021 37.26995 64.43633 81.87090 91.01148 2
#> 2: 2021/2022 40.28893 66.68661 82.29283 90.30798 3
#> 3: 2022/2023 41.83822 71.02438 81.28250 86.27663 4
#> weeks weeks_with_no_threshold
#> 212 104
#> isoyearweek numerator_nowcasted_status_prob_preepidemic
#> <char> <num>
#> 1: 2021-45 1.00
#> 2: 2022-02 0.02
#> numerator_nowcasted_status_prob_low numerator_nowcasted_status_prob_medium
#> <num> <num>
#> 1: 0.00 0.00
#> 2: 0.94 0.04
#> numerator_nowcasted_status_prob_high
#> <num>
#> 1: 0
#> 2: 0
#> numerator_nowcasted_status_prob_veryhigh
#> <num>
#> 1: 0
#> 2: 0