Package 'hubEvals'

Score model output predictions

Description

Scores model outputs with a single output_type against observed data.

Usage

score_model_out(
  model_out_tbl,
  oracle_output,
  metrics = NULL,
  relative_metrics = NULL,
  baseline = NULL,
  summarize = TRUE,
  by = "model_id",
  output_type_id_order = NULL,
  compound_taskid_set = NULL,
  transform = NULL,
  transform_append = FALSE,
  transform_label = NULL,
  ...
)

Arguments

model_out_tbl	Model output tibble with predictions
oracle_output	Predictions that would have been generated by an oracle model that knew the observed target data values in advance
metrics	Character vector of scoring metrics to compute. If NULL (the default), appropriate metrics are chosen automatically. See details for more.
relative_metrics	Character vector of scoring metrics for which to compute relative skill scores. The relative_metrics should be a subset of metrics and should only include proper scores (e.g., it should not contain interval coverage metrics). If NULL (the default), no relative metrics will be computed. Relative metrics are only computed if summarize = TRUE, and require that "model_id" is included in by.
baseline	String with the name of a model to use as a baseline for relative skill scores. If a baseline is given, then a scaled relative skill with respect to the baseline will be returned. By default (NULL), relative skill will not be scaled with respect to a baseline model.
summarize	Boolean indicator of whether summaries of forecast scores should be computed. Defaults to TRUE.
by	Character vector naming columns to summarize by. For example, specifying by = "model_id" (the default) will compute average scores for each model.
output_type_id_order	For ordinal variables in pmf format, this is a vector of levels for pmf forecasts, in increasing order of the levels. The order of the values for the output_type_id can be found by referencing the hub's tasks.json configuration file. For all output types other than pmf, this is ignored.
compound_taskid_set	When NULL (the default), sample forecasts are scored marginally (each modeling task scored independently). When a character vector of task ID column names is provided, it sets the compound grouping: the specified columns stay constant within each sample draw, while the remaining task ID dimensions vary within a draw and are scored jointly (using the energy score). Only applicable when output_type == "sample". The value of compound_taskid_set can be found by referencing the hub's tasks.json configuration file.
transform	A function to apply as a scale transformation to both predictions and observations before scoring. Common choices include log_shift (recommended for log transformation as it handles zeros via an offset parameter), sqrt, or log1p. Avoid using log directly if data may contain zeros. If NULL (the default), no transformation is applied. Only supported for quantile, mean, and median output types.
transform_append	Logical. If FALSE (the default), scores are computed only on the transformed scale. If TRUE, scores are computed on both original and transformed scales, with a scale column distinguishing them. Ignored if transform = NULL.
transform_label	A character string label for the transformation (e.g., "log"). If NULL (the default), the label is auto-generated from the function name (e.g., "log_shift" for scoringutils::log_shift). Required when using an anonymous transform function. Ignored if transform = NULL. Note: the label only appears in output when transform_append = TRUE, where it distinguishes transformed rows (labeled with this value) from original rows (labeled "natural") in the scale column.
...	Additional arguments passed to the transform function. For example, allows use of the offset and base arguments of scoringutils::log_shift(). Ignored if transform = NULL.

Details

See the hubverse documentation for the expected format of the oracle output data.

Default metrics are provided by the scoringutils package. You can select metrics by passing in a character vector of metric names to the metrics argument.

The following metrics can be selected (all are used by default) for the different output_types:

Quantile forecasts: (output_type == "quantile")

• wis

• overprediction

• underprediction

• dispersion

• bias

• ae_median

• "interval_coverage_XX": interval coverage at the "XX" level. For example, "interval_coverage_95" is the 95% interval coverage rate, which would be calculated based on quantiles at the probability levels 0.025 and 0.975.

See scoringutils::get_metrics.forecast_quantile for details.

Nominal forecasts: (output_type == "pmf" and output_type_id_order is NULL)

• log_score

See scoringutils::get_metrics.forecast_nominal for details.

Ordinal forecasts: (output_type == "pmf" and output_type_id_order is a vector)

• log_score

• rps

See scoringutils::get_metrics.forecast_ordinal for details.

Median forecasts: (output_type == "median")

• ae_point: absolute error of the point forecast (recommended for the median, see Gneiting (2011))

See scoringutils::get_metrics.forecast_point for details.

Mean forecasts: (output_type == "mean")

• se_point: squared error of the point forecast (recommended for the mean, see Gneiting (2011))

Sample forecasts (marginal): (output_type == "sample", compound_taskid_set = NULL)

• bias

• dss

• crps

• overprediction

• underprediction

• dispersion

• log_score

• mad

• ae_median

• se_mean

Note: log_score uses kernel density estimation, which may not be appropriate for integer-valued forecasts. scoringutils will warn when this is detected.

See scoringutils::get_metrics.forecast_sample for details.

Sample forecasts (compound): (output_type == "sample", compound_taskid_set provided)

• energy_score

• variogram_score

See scoringutils::get_metrics.forecast_sample_multivariate for details. The output includes a .mv_group_id column assigned by scoringutils to identify the multivariate groups used for scoring (equivalent to the compound_idx concept in the hubverse sample output type documentation ). Correct scoring depends on providing the right compound_taskid_set from the hub's tasks.json configuration. If the specified grouping does not match the actual dependence structure of the submitted samples (e.g., because some models submitted coarser samples than configured), .mv_group_id may not correspond to the original sample draws as indicated by their output_type_id values.

See scoringutils::add_relative_skill for details on relative skill scores.

Value

A data.table with scores

References

Gneiting, Tilmann. 2011. "Making and Evaluating Point Forecasts." Journal of the American Statistical Association 106 (494): 746–62. <doi: 10.1198/jasa.2011.r10138>.

Examples

# compute WIS and interval coverage rates at 80% and 90% levels based on
# quantile forecasts, summarized by the mean score for each model
quantile_scores <- score_model_out(
  model_out_tbl = hubExamples::forecast_outputs |>
    dplyr::filter(.data[["output_type"]] == "quantile"),
  oracle_output = hubExamples::forecast_oracle_output,
  metrics = c("wis", "interval_coverage_80", "interval_coverage_90"),
  relative_metrics = "wis",
  by = "model_id"
)
quantile_scores

# compute log scores based on pmf predictions for categorical targets,
# summarized by the mean score for each combination of model and location.
# Note: if the model_out_tbl had forecasts for multiple targets using a
# pmf output_type with different bins, it would be necessary to score the
# predictions for those targets separately.
pmf_scores <- score_model_out(
  model_out_tbl = hubExamples::forecast_outputs |>
    dplyr::filter(.data[["output_type"]] == "pmf"),
  oracle_output = hubExamples::forecast_oracle_output,
  metrics = c("log_score", "rps"),
  by = c("model_id", "location", "horizon"),
  output_type_id_order = c("low", "moderate", "high", "very high")
)
head(pmf_scores)

# Score sample forecasts marginally (each modeling task scored independently).
# Note: this data has compound structure (samples span horizons), but marginal
# scoring is still valid -- it evaluates each horizon independently.
sample_scores <- score_model_out(
  model_out_tbl = hubExamples::forecast_outputs |>
    dplyr::filter(.data[["output_type"]] == "sample"),
  oracle_output = hubExamples::forecast_oracle_output,
  metrics = "crps",
  by = "model_id"
)
sample_scores

# Score compound sample forecasts jointly using the energy score.
# compound_taskid_set specifies which task IDs stay constant within
# a sample group -- here, each sample draw spans all horizons for a
# given reference_date and location (i.e., a trajectory over time).
compound_scores <- score_model_out(
  model_out_tbl = hubExamples::forecast_outputs |>
    dplyr::filter(.data[["output_type"]] == "sample"),
  oracle_output = hubExamples::forecast_oracle_output,
  compound_taskid_set = c("reference_date", "location"),
  by = "model_id"
)
compound_scores

---

Transform pmf model output into a forecast object

Description

Transform pmf model output into a forecast object

Usage

transform_pmf_model_out(
  model_out_tbl,
  oracle_output,
  output_type_id_order = NULL
)

Arguments

model_out_tbl	Model output tibble with predictions
oracle_output	Predictions that would have been generated by an oracle model that knew the observed target data values in advance
output_type_id_order	For ordinal variables in pmf format, this is a vector of levels for pmf forecasts, in increasing order of the levels. The order of the values for the output_type_id can be found by referencing the hub's tasks.json configuration file. For all output types other than pmf, this is ignored.

Value

forecast_quantile

---

Transform either mean or median model output into a point forecast object:

Description

Transform either mean or median model output into a point forecast object:

Usage

transform_point_model_out(model_out_tbl, oracle_output, output_type)

Arguments

model_out_tbl	Model output tibble with predictions
oracle_output	Predictions that would have been generated by an oracle model that knew the observed target data values in advance
output_type	Forecast output type: "mean" or "median"

Details

This function transforms a model output tibble in the Hubverse format (with either "mean" or "median" output type) to a scoringutils "point" forecast object

Value

forecast_point

---

Transform quantile model output into a forecast object

Description

Transform quantile model output into a forecast object

Usage

transform_quantile_model_out(model_out_tbl, oracle_output)

Arguments

model_out_tbl	Model output tibble with predictions
oracle_output	Predictions that would have been generated by an oracle model that knew the observed target data values in advance

Value

forecast_quantile

---

Transform sample model output into a forecast object

Description

Transform sample model output into a forecast object

Usage

transform_sample_model_out(
  model_out_tbl,
  oracle_output,
  compound_taskid_set = NULL
)

Arguments

model_out_tbl	Model output tibble with predictions
oracle_output	Predictions that would have been generated by an oracle model that knew the observed target data values in advance
compound_taskid_set	Character vector of task ID column names that stay constant within each sample draw (i.e., define the compound modeling task grouping). When NULL (the default), each modeling task is scored independently (marginal scoring). When provided, sample draws are treated as joint predictions across the task ID dimensions not in compound_taskid_set, and multivariate scoring metrics are used.

Value

A forecast_sample object (when compound_taskid_set is NULL) or a forecast_sample_multivariate object (when compound_taskid_set is provided).

---