eval-harness

dataset

ag_news_20

20 items · 0 arms. Headline: each arm's own primary metric, mean over its items; arms with different primary metrics are not ranked against each other.

Open its sheet

goals: News topic classification (AG News)

01arms

scored against silverscored by a judgewrote the reference it is scored againstscored against a label set, not a reference setunmarked rows are scored against gold

0 shown
armmetricmeanitemscost per itemspeed per itemrunsscored against

02what's inside

20 items, text, 24–136 words (median 47, middle half 35–55).

Source: fancyzhx/ag_news (test split). The text stays there; each item below links to its row.

051024–35 words: 636–47 words: 548–59 words: 460–71 words: 4132–143 words: 124–3536–4748–5960–7172–8384–9596–107108–119120–131132–143

the gold

gold/ag_news_20: 4 labels

labelitemsshare
Business525%
Sci/Tech525%
Sports525%
World525%
20 shown
itemwordsgoldopen
013ffb81b47f25Businesssource ↗
0c47eb9eeb6c67Sportssource ↗
0e7de46882b269Sportssource ↗
108a71f66723136Sci/Techsource ↗
2510d175de8d70Sportssource ↗
416fc273495b24Worldsource ↗
655d89d9592431Sportssource ↗
7e2d0886edef45Sci/Techsource ↗
8df82604abfc41Businesssource ↗
a1164f3023b048Sci/Techsource ↗
a787b831978f44Worldsource ↗
afbd779348f030Businesssource ↗
bee7c037baed47Businesssource ↗
c2d5dcb0c46934Sci/Techsource ↗
cbdcc8db300335Sportssource ↗
d2ddffdd6e9055Sci/Techsource ↗
d8817a3ea9e146Businesssource ↗
e1d8311be5d952Worldsource ↗
e35f3ecb1e0255Worldsource ↗
eced564834db64Worldsource ↗

Each row is the profile evh dataset-items --dataset-id ag_news_20 prints; inspect opens the item in the inspector with the dataset's largest group of comparable arms, and its text where the inspector runs on a machine that has it.

03reference sets

04labels

No labels yet. A sheet to fill: evh label-sheet --dataset-id ag_news_20 --out sheet.csv; a run's outputs as a model set: evh label-from-run --run R --set S.