Datasets:
Tasks:
Text Classification
Formats:
parquet
Languages:
English
Size:
1K - 10K
Tags:
structured-decisions
calibration
probabilistic-classification
system-one
workflow-evaluation
Synthetic
License:
Add eval.yaml to register as a Hub benchmark
#9
by SaylorTwift HF Staff - opened
eval.yaml
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
name: Typed Decisions
|
| 2 |
+
description: >
|
| 3 |
+
A benchmark for typed probabilistic decisions. A model gets one piece of
|
| 4 |
+
unstructured state and answers five typed questions about it at once, and
|
| 5 |
+
every answer is a probability distribution, not a single label.
|
| 6 |
+
|
| 7 |
+
tasks:
|
| 8 |
+
- id: accuracy
|
| 9 |
+
config: all
|
| 10 |
+
split: test
|
| 11 |
+
- id: kl_from_gold
|
| 12 |
+
config: all
|
| 13 |
+
split: test
|
| 14 |
+
lower_is_better: true
|
| 15 |
+
- id: brier
|
| 16 |
+
config: all
|
| 17 |
+
split: test
|
| 18 |
+
lower_is_better: true
|
| 19 |
+
- id: ece
|
| 20 |
+
config: all
|
| 21 |
+
split: test
|
| 22 |
+
lower_is_better: true
|