Takes a list of times, classes and counts, e.g. a COGUK variant like data set
with time, (multinomial) class (e.g. variant) and count being the count in
that time period. Fits a quadratic B-spline on time to the proportion of the
data using nnet::multinom, with approx one degree of freedom per class and per
window units of the time series.
Usage
multinomial_nnet_model(
d = i_multinomial_input,
...,
window = 14,
frequency = "1 day",
predict = TRUE,
.progress = interactive()
)Arguments
- d
A multi-class count input dataframe - a dataframe with columns:
class (factor) - A factor specifying the type of observation. This will be things like variant, or serotype, for a multinomial model. Any missing data points are ignored.
count (positive_integer) - Positive case counts associated with the specified time frame
time (ggoutbreak::time_period + group_unique) - A (usually complete) set of singular observations per unit time as a `time_period`
Minimally grouped by: class (and other groupings allowed).
- ...
not used
- window
a number of data points between knots, smaller values result in less smoothing, large value in more.
- frequency
the temporal density of the output estimates.
- predict
result a prediction. If false we return the model.
- .progress
show a CLI progress bar
Examples
data = test_poisson_rt_2class
tmp = data %>% ggoutbreak::multinomial_nnet_model(window=14)
#> # weights: 28 (13 variable)
#> initial value 43356.356144
#> iter 10 value 38674.268766
#> iter 20 value 37400.259934
#> final value 37399.822194
#> converged
if (interactive()) {
plot_multinomial(tmp, date_labels="%b %y")
}