quantile_example_data_wide

A toy dataset for quantile forecasts of an outcome variable

datasets

Combines a collection of metrics and proper scoring rules
(Tilmann Gneiting & Adrian E Raftery (2007)
<doi:10.1198/016214506000001437>) with an easy to
use wrapper that can be used to automatically evaluate predictions.
Apart from proper scoring rules functions are provided to assess bias,
sharpness and calibration (Sebastian Funk, Anton Camacho, Adam J. Kucharski,
Rachel Lowe, Rosalind M. Eggo, W. John Edmunds (2019)
<doi:10.1371/journal.pcbi.1006785>) of forecasts.
Several types of predictions can be evaluated:
probabilistic forecasts (generally
predictive samples generated by Markov Chain Monte Carlo procedures),
quantile forecasts or point forecasts. Observed values and predictions
can be either continuous, integer, or binary. Users can either choose
to apply these rules separately in a vector / matrix format that can
be flexibly used within other packages, or they can choose to do an
automatic evaluation of their forecasts. This is implemented with
'data.table' and provides a consistent and very efficient framework for
evaluating various types of predictions.

Nikos Bosse

scoringutils

Utilities for Scoring and Assessing Predictions

Sam Abbott 

Joel Hellewell

Sophie Meakins 

James Munday

Katharine Sherratt

Sebastian Funk

quantile_example_data_wide function

A data frame with 120 rows and 10 variables:<dl class="dl-horizontal">
 <dt>true_value</dt><dd>true observed values</dd>
 <dt>id</dt><dd>unique identifier for true observed values</dd>
 <dt>model</dt><dd>name of the model that generated the forecasts</dd>
 <dt>horizon</dt><dd>forecast horzin (e.g. 1 day ahead forecast)</dd>
 <dt>lower_90</dt><dd>prediction for the lower value of the 90% interval range
 (corresponding to the 5% quantile)</dd>
 <dt>lower_50</dt><dd>prediction for the lower value of the 50% interval range
 (corresponding to the 25% quantile)</dd>
 <dt>lower_0</dt><dd>prediction for the lower value of the 0% interval range
 (corresponding to the 50% quantile, i.e. the median. For computational
 reasons there need be a column with <code>lower_0</code> and <code>upper_0</code>)</dd>
 <dt>upper_0</dt><dd>prediction for the upper value of the 0
 (corresponding to the 50% quantile, i.e. the median)</dd>
 <dt>upper_50</dt><dd>prediction for the upper value of the 50% interval range
 (corresponding to the 75% quantile)</dd>
 <dt>upper_90</dt><dd>prediction for the lower value of the 90% interval range
 (corresponding to the 95% quantile)</dd>
</dl>

Format

Quantile Example Data - Wide Format — quantile_example_data_wide

A data frame with 120 rows and 10 variables:<dl class='dl-horizontal'>
 <dt>true_value</dt><dd>true observed values</dd>
 <dt>id</dt><dd>unique identifier for true observed values</dd>
 <dt>model</dt><dd>name of the model that generated the forecasts</dd>
 <dt>horizon</dt><dd>forecast horzin (e.g. 1 day ahead forecast)</dd>
 <dt>lower_90</dt><dd>prediction for the lower value of the 90% interval range
 (corresponding to the 5% quantile)</dd>
 <dt>lower_50</dt><dd>prediction for the lower value of the 50% interval range
 (corresponding to the 25% quantile)</dd>
 <dt>lower_0</dt><dd>prediction for the lower value of the 0% interval range
 (corresponding to the 50% quantile, i.e. the median. For computational
 reasons there need be a column with <code>lower_0</code> and <code>upper_0</code>)</dd>
 <dt>upper_0</dt><dd>prediction for the upper value of the 0
 (corresponding to the 50% quantile, i.e. the median)</dd>
 <dt>upper_50</dt><dd>prediction for the upper value of the 50% interval range
 (corresponding to the 75% quantile)</dd>
 <dt>upper_90</dt><dd>prediction for the lower value of the 90% interval range
 (corresponding to the 95% quantile)</dd>
</dl>

quantile_example_data_wide: Quantile Example Data - Wide Format

Description

Usage

Arguments

Format