Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -442,3 +442,24 @@ configs:
|
|
| 442 |
- split: test
|
| 443 |
path: two_ev-graham/test-*
|
| 444 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 442 |
- split: test
|
| 443 |
path: two_ev-graham/test-*
|
| 444 |
---
|
| 445 |
+
|
| 446 |
+
# Evaluation code:
|
| 447 |
+
```python
|
| 448 |
+
|
| 449 |
+
def parse(x):
|
| 450 |
+
if '<answer>' in x and '</answer>' in x:
|
| 451 |
+
start = x.find('<answer>') + len('<answer>')
|
| 452 |
+
end = x.find('</answer>')
|
| 453 |
+
x = x[start:end]
|
| 454 |
+
|
| 455 |
+
lines = [i.lstrip('L').strip() for i in x.strip().strip('.').split(',')]
|
| 456 |
+
return [int(i) for i in lines if i.isnumeric()]
|
| 457 |
+
|
| 458 |
+
def jaccard(list1, list2):
|
| 459 |
+
intersection = len(list(set(list1).intersection(list2)))
|
| 460 |
+
union = (len(set(list1)) + len(set(list2))) - intersection
|
| 461 |
+
return float(intersection) / union
|
| 462 |
+
|
| 463 |
+
df['yp']=df.prompt.parallel_apply(complete_m(m,num_retries=5))
|
| 464 |
+
df['metric']=df.apply(lambda x: jaccard(parse(x.yp), x.evidence_indices), axis=1)
|
| 465 |
+
```
|