Initialize predictions companion for owid-mpox
Browse files
README.md
ADDED
|
@@ -0,0 +1,72 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
pretty_name: "Predictions \u2014 OWID Mpox \u2014 global daily compiled"
|
| 3 |
+
license: other
|
| 4 |
+
tags:
|
| 5 |
+
- epi-eval
|
| 6 |
+
- predictions
|
| 7 |
+
- forecast-evaluation
|
| 8 |
+
- companion-of-owid-mpox
|
| 9 |
+
configs:
|
| 10 |
+
- config_name: default
|
| 11 |
+
data_files:
|
| 12 |
+
- split: train
|
| 13 |
+
path: data/*.parquet
|
| 14 |
+
---
|
| 15 |
+
|
| 16 |
+
# Predictions for OWID Mpox — global daily compiled
|
| 17 |
+
|
| 18 |
+
Community-submitted forecasts targeting [`EPI-Eval/owid-mpox`](https://huggingface.co/datasets/EPI-Eval/owid-mpox).
|
| 19 |
+
Each row is one quantile (or point) forecast for one target date — see the
|
| 20 |
+
schema below.
|
| 21 |
+
|
| 22 |
+
This repo accumulates accepted submissions from many forecasters. New
|
| 23 |
+
predictions arrive as community pull requests opened from the [EPI-Eval
|
| 24 |
+
dashboard](https://github.com/ChrisHarig/apart-forecasting-tool); a
|
| 25 |
+
maintainer reviews each PR before merging.
|
| 26 |
+
|
| 27 |
+
## Schema (v1)
|
| 28 |
+
|
| 29 |
+
| column | type | notes |
|
| 30 |
+
| --- | --- | --- |
|
| 31 |
+
| `target_date` | string (`YYYY-MM-DD`) | The date being forecast |
|
| 32 |
+
| `target_dataset` | string | Always `owid-mpox` |
|
| 33 |
+
| `target_column` | string | Truth column being forecast (see below) |
|
| 34 |
+
| `submitter` | string | Forecaster name or HF username |
|
| 35 |
+
| `model_name` | string | Identifier for the model run |
|
| 36 |
+
| `description` | string | Free-form notes on the model |
|
| 37 |
+
| `quantile` | float (nullable) | In `[0, 1]`. `null` = point estimate |
|
| 38 |
+
| `value` | float | Forecast value (in the truth column's units) |
|
| 39 |
+
| `submitted_at` | string (ISO 8601) | UTC submission timestamp |
|
| 40 |
+
| _(pass-through dims)_ | string | Categorical dims from the source CSV |
|
| 41 |
+
|
| 42 |
+
Long format: one row per `(target_date, [dim values…], quantile)`. A
|
| 43 |
+
forecaster providing the median plus 50%/80%/95% intervals emits 7 rows per
|
| 44 |
+
date (one point + 6 quantiles). Multiple submissions from the same forecaster
|
| 45 |
+
land as separate parquet files under `data/`.
|
| 46 |
+
|
| 47 |
+
## Forecast targets
|
| 48 |
+
|
| 49 |
+
Truth columns from `EPI-Eval/owid-mpox` you can forecast:
|
| 50 |
+
|
| 51 |
+
- `new_cases` (cases/day)
|
| 52 |
+
- `new_deaths` (deaths/day)
|
| 53 |
+
- `total_cases` (cases (cumulative))
|
| 54 |
+
- `total_deaths` (deaths (cumulative))
|
| 55 |
+
|
| 56 |
+
## Submitting
|
| 57 |
+
|
| 58 |
+
The dashboard at [apart-forecasting-tool](https://github.com/ChrisHarig/apart-forecasting-tool)
|
| 59 |
+
handles the full submission flow: drag-drop a CSV, pick this dataset as the
|
| 60 |
+
"Compare to" target, review your scores against the truth, and click "Submit
|
| 61 |
+
to HuggingFace." The dashboard serializes your CSV into the schema above and
|
| 62 |
+
opens a community PR here.
|
| 63 |
+
|
| 64 |
+
## Notes
|
| 65 |
+
|
| 66 |
+
- Predictions whose `target_date` falls outside the truth dataset's coverage
|
| 67 |
+
(forecast horizon) are still accepted. Comparison metrics on the dashboard
|
| 68 |
+
compute only on the dates where truth is available.
|
| 69 |
+
- A submission's `submitter` value is its only identity claim — there's no
|
| 70 |
+
signed authentication. Reviewers should sanity-check unfamiliar submitters.
|
| 71 |
+
|
| 72 |
+
_Initialized by `upload_pipeline.core.bootstrap_predictions_repos`._
|