EnVar microschema · class
Model Aggregate Uncertainty
ModelAggregateUncertainty
Whole-model uncertainty summary — cross-validation metrics and the reference where they are reported. Inlined on model_aggregate_uncertainty.
Where it sits
Composed intoUncertainty
Fields
A 0-to-1 score of how well the model's predictions matched reality on data it was not trained on — 1 is perfect, 0 is no better than always guessing the average. "Cross-validated" means the test used held-out data, so the score is honest.
Cross-validated R² for the model as a whole.
Example
0.86 — Di et al. ensemble PM2.5
Why it matters & mappings
The single most comparable headline number for model skill. Without it a consumer cannot weigh this product against an alternative, or decide whether the model is good enough for the health analysis at hand.
Root-mean-square error — the typical size of the model's mistakes, in the same units as the value itself (e.g. °C or µg/m³). Smaller is better.
Cross-validated RMSE for the model as a whole.
Example
1.89 — Di et al. ensemble PM2.5, in µg/m³ (same model as the cv_r2 example)
Why it matters & mappings
R² alone hides how large the errors actually are. RMSE states the typical error in the value's own units, which is what determines whether model error is negligible or fatal for a given effect size.
Where these numbers come from — the paper or report (ideally a DOI) so a reader can check them at the source.
DOI / citation where the aggregate uncertainty is reported.
Example
10.1016/j.envint.2019.104909 — DOI of the methods paper reporting the metrics
Why it matters & mappings
Uncertainty numbers copied into a sidecar are only as trustworthy as their source. Without the reference the metrics cannot be verified, attributed, or updated when the producer revises them.
Full field reference — every slot, cardinality & inheritance
| Field | Name | Tier | Cardinality / Range | Description |
|---|---|---|---|---|
| Cross-Validated R² | cv_r2 |
optional | 0..1 Float |
Cross-validated R² for the model as a whole |
| Cross-Validated RMSE | cv_rmse |
optional | 0..1 Float |
Cross-validated RMSE for the model as a whole |
| Reporting Reference | reported_in |
optional | 0..1 String |
DOI / citation where the aggregate uncertainty is reported |
Diagram & LinkML source
classDiagram
class ModelAggregateUncertainty
click ModelAggregateUncertainty href "../../classes/ModelAggregateUncertainty/"
ModelAggregateUncertainty : cv_r2
ModelAggregateUncertainty : cv_rmse
ModelAggregateUncertainty : reported_in
name: ModelAggregateUncertainty
description: Whole-model uncertainty summary — cross-validation metrics and the reference
where they are reported. Inlined on `model_aggregate_uncertainty`.
title: Model Aggregate Uncertainty
from_schema: https://w3id.org/linkml/microschemas/envar
see_also:
- https://en.wikipedia.org/wiki/Root_mean_square_deviation
- https://en.wikipedia.org/wiki/Coefficient_of_determination
rank: 1000
attributes:
cv_r2:
name: cv_r2
annotations:
justification:
tag: justification
value: The single most comparable headline number for model skill. Without
it a consumer cannot weigh this product against an alternative, or decide
whether the model is good enough for the health analysis at hand.
explanation:
tag: explanation
value: A 0-to-1 score of how well the model's predictions matched reality
on data it was not trained on — 1 is perfect, 0 is no better than always
guessing the average. "Cross-validated" means the test used held-out data,
so the score is honest.
description: Cross-validated R² for the model as a whole.
title: Cross-Validated R²
examples:
- value: '0.86'
description: Di et al. ensemble PM2.5
from_schema: https://w3id.org/linkml/microschemas/envar/uncertainty
rank: 1000
owner: ModelAggregateUncertainty
domain_of:
- ModelAggregateUncertainty
range: float
cv_rmse:
name: cv_rmse
annotations:
justification:
tag: justification
value: R² alone hides how large the errors actually are. RMSE states the typical
error in the value's own units, which is what determines whether model error
is negligible or fatal for a given effect size.
explanation:
tag: explanation
value: Root-mean-square error — the typical size of the model's mistakes,
in the same units as the value itself (e.g. °C or µg/m³). Smaller is better.
description: Cross-validated RMSE for the model as a whole.
title: Cross-Validated RMSE
examples:
- value: '1.89'
description: Di et al. ensemble PM2.5, in µg/m³ (same model as the cv_r2 example)
from_schema: https://w3id.org/linkml/microschemas/envar/uncertainty
rank: 1000
owner: ModelAggregateUncertainty
domain_of:
- ModelAggregateUncertainty
range: float
reported_in:
name: reported_in
annotations:
justification:
tag: justification
value: Uncertainty numbers copied into a sidecar are only as trustworthy as
their source. Without the reference the metrics cannot be verified, attributed,
or updated when the producer revises them.
explanation:
tag: explanation
value: Where these numbers come from — the paper or report (ideally a DOI)
so a reader can check them at the source.
description: DOI / citation where the aggregate uncertainty is reported.
title: Reporting Reference
examples:
- value: 10.1016/j.envint.2019.104909
description: DOI of the methods paper reporting the metrics
from_schema: https://w3id.org/linkml/microschemas/envar/uncertainty
rank: 1000
owner: ModelAggregateUncertainty
domain_of:
- ModelAggregateUncertainty
range: string
See Also
- https://en.wikipedia.org/wiki/Root_mean_square_deviation
- https://en.wikipedia.org/wiki/Coefficient_of_determination
Identifier and Mapping Information
Schema Source
- from schema: https://w3id.org/linkml/microschemas/envar
Mappings
| Mapping Type | Mapped Value |
|---|---|
| self | envar:ModelAggregateUncertainty |
| native | envar:ModelAggregateUncertainty |