This PR closes #203. It uses v2.0.0 of scoringutils to compute the full set of scores based on the samples and quantiles of the forecasted hospital admissions in the evaluation pipeline.
Note this is a WIP. Currently we have just edited the inputs to scoringutils::score() which expects new names for the predicted and observed values and requires converting from a data.table to a forecast object. For compatibility with the rest of the downstream functions (mostly plotting) I converted names back to what we had previously named the columns (true_value and prediction). This might be incompatible with summarise_scores().
To do
[ ] test eval_postprocess() locally on a ww_fit_obj and ensure the scores are equivalent or nearly to the previous implementation
[ ] run on full _targets_eval_postprocessing.R or targets_subset_benchmarking.R to ensure compatibility with downstream scoringutils functions
This PR closes #203. It uses
v2.0.0
ofscoringutils
to compute the full set of scores based on the samples and quantiles of the forecasted hospital admissions in the evaluation pipeline.Note this is a WIP. Currently we have just edited the inputs to
scoringutils::score()
which expects new names for the predicted and observed values and requires converting from adata.table
to a forecast object. For compatibility with the rest of the downstream functions (mostly plotting) I converted names back to what we had previously named the columns (true_value
andprediction
). This might be incompatible withsummarise_scores()
.To do
eval_postprocess()
locally on aww_fit_obj
and ensure the scores are equivalent or nearly to the previous implementation_targets_eval_postprocessing.R
ortargets_subset_benchmarking.R
to ensure compatibility with downstreamscoringutils
functions