Forecasting Swiss ILI counts using kcde::kcde

Sebastian Meyer

2023-11-29

Source: vignettes/extra/CHILI_kcde.Rmd

In this vignette, we use forecasting methods provided by:

The corresponding software reference is:

Ray E (2017). kcde: Kernel conditional density estimation with flexible kernel specifications. R package version 0.0.0.9000, https://github.com/reichlab/kcde/tree/6f8c8e5c82ec63e2e8f8bc2a143ec75460d86ab4.

Modelling

Note: we use a log-transformation of the CHILI counts in kcde.

Configuring kcde() is quiet lengthy and not shown here (see the vignette sources for details).

We used 3 cores in parallel. Fitting the full bandwidth KCDE would take several days, so we used the diagonal bandwidth parametrization.

kcdefit$runtime / 60
##    user  system elapsed 
## 19.3431  0.4708  6.9494

Unfortunately, the estimation function kcde() just returns a list without a dedicated class. It is unclear to me how to summarize the model fit or extract fitted values …

One-week-ahead forecasts

We compute 213 one-week-ahead forecasts from 2012-W48 to 2016-W51 (the OWA period).

The code is again lengthy and not shown here (see the sources for details).

Computing the 213 forecasts took 6.5 minutes (single core).

The PIT histogram is based on the pointwise ECDF of the samples from the predictive distributions.

##       DSS          log_score           AE       
##  Min.   : 8.61   Min.   : 4.50   Min.   :    1  
##  1st Qu.:11.02   1st Qu.: 6.52   1st Qu.:   99  
##  Median :13.27   Median : 7.64   Median :  340  
##  Mean   :13.79   Mean   : 7.80   Mean   :  956  
##  3rd Qu.:16.85   3rd Qu.: 9.09   3rd Qu.:  910  
##  Max.   :23.36   Max.   :13.01   Max.   :10872

Long-term forecasts

We would need to rerun kcde() and subsequent predictions for each of the three different training periods with prediction_horizon varying from 1 to 30. Based on the runtime of the above computations for a single training period and prediction horizon, computing all long-term forecasts is estimated to take approximately 20.2 hours. This is far beyond the runtimes of the alternative prediction approaches, so we skip long-term forecasts with kcde.