File size: 1,498 Bytes
02b83fa
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
"""Extract linguistic + math feature matrices for ASDiv (grade-labeled) and SVAMP."""
import json
import pandas as pd
from datasets import load_dataset
from features import linguistic_features, math_features_from_chain, math_features_from_equation


def build_asdiv():
    ds = load_dataset("MU-NLPC/Calc-asdiv_a", split="test")
    rows = []
    for ex in ds:
        q = ex["question"]
        if not q or not q.strip():
            continue
        lf = linguistic_features(q)
        mf = math_features_from_chain(ex["chain"] or "")
        row = {"id": ex["id"], "grade": ex["grade"], "text": q}
        row.update(lf)
        row.update(mf)
        rows.append(row)
    df = pd.DataFrame(rows)
    df.to_parquet("asdiv_features.parquet")
    print("ASDiv:", df.shape, "grades:", sorted(df.grade.unique()))
    print(df.grade.value_counts().sort_index().to_dict())
    return df


def build_svamp():
    ds = load_dataset("ChilleD/SVAMP", split="train")
    rows = []
    for ex in ds:
        text = (ex["Body"] + " " + ex["Question"]).strip()
        lf = linguistic_features(text)
        mf = math_features_from_equation(ex["Equation"] or "")
        row = {"id": ex["ID"], "type": ex["Type"], "text": text}
        row.update(lf)
        row.update(mf)
        rows.append(row)
    df = pd.DataFrame(rows)
    df.to_parquet("svamp_features.parquet")
    print("SVAMP:", df.shape, "types:", df.type.nunique())
    return df


if __name__ == "__main__":
    build_asdiv()
    build_svamp()