Spaces:
Running
Running
Create mcqa_dataset.py
Browse files- mcqa_dataset.py +55 -0
mcqa_dataset.py
ADDED
|
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# mcqa_dataset.py
|
| 2 |
+
# --------------------------------------------------
|
| 3 |
+
# Pre‑tokenised dataset for 4‑choice MCQA
|
| 4 |
+
# --------------------------------------------------
|
| 5 |
+
import json
|
| 6 |
+
import torch
|
| 7 |
+
from torch.utils.data import Dataset
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
class MCQADataset(Dataset):
|
| 11 |
+
"""
|
| 12 |
+
Each item returns:
|
| 13 |
+
input_ids, attention_mask : LongTensor (max_len)
|
| 14 |
+
label : 0/1 (1 → correct choice)
|
| 15 |
+
qid, cid : strings (question id, choice id)
|
| 16 |
+
"""
|
| 17 |
+
def __init__(self, path: str, tokenizer, max_len: int = 128):
|
| 18 |
+
self.encodings, self.labels, self.qids, self.cids = [], [], [], []
|
| 19 |
+
|
| 20 |
+
with open(path, encoding="utf-8") as f:
|
| 21 |
+
for line in f:
|
| 22 |
+
obj = json.loads(line)
|
| 23 |
+
stem = obj["question"]["stem"]
|
| 24 |
+
fact = obj["fact1"]
|
| 25 |
+
gold = obj["answerKey"]
|
| 26 |
+
|
| 27 |
+
for ch in obj["question"]["choices"]:
|
| 28 |
+
text = f"{fact} {stem} {ch['text']}"
|
| 29 |
+
enc = tokenizer(
|
| 30 |
+
text,
|
| 31 |
+
max_length=max_len,
|
| 32 |
+
truncation=True,
|
| 33 |
+
padding="max_length",
|
| 34 |
+
)
|
| 35 |
+
self.encodings.append(enc)
|
| 36 |
+
self.labels.append(1 if ch["label"] == gold else 0)
|
| 37 |
+
self.qids.append(obj["id"])
|
| 38 |
+
self.cids.append(ch["label"])
|
| 39 |
+
|
| 40 |
+
# Convert lists of dicts → dict of lists for cheaper indexing
|
| 41 |
+
self.encodings = {
|
| 42 |
+
k: [d[k] for d in self.encodings] for k in self.encodings[0]
|
| 43 |
+
}
|
| 44 |
+
|
| 45 |
+
# --------------------------------------------------
|
| 46 |
+
|
| 47 |
+
def __len__(self):
|
| 48 |
+
return len(self.labels)
|
| 49 |
+
|
| 50 |
+
def __getitem__(self, idx):
|
| 51 |
+
item = {k: torch.tensor(v[idx]) for k, v in self.encodings.items()}
|
| 52 |
+
item["label"] = torch.tensor(self.labels[idx], dtype=torch.long)
|
| 53 |
+
item["qid"] = self.qids[idx]
|
| 54 |
+
item["cid"] = self.cids[idx]
|
| 55 |
+
return item
|