pietrolesci
commited on
Commit
•
8371e5c
1
Parent(s):
ef4ea9a
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
## Overview
|
2 |
+
|
3 |
+
Original dataset available [here](https://github.com/sheng-z/JOCI/tree/master/data).
|
4 |
+
This dataset is the "full" JOCI dataset, which is the file named `joci.csv.zip`.
|
5 |
+
|
6 |
+
|
7 |
+
# Dataset curation
|
8 |
+
The following processing is applied,
|
9 |
+
|
10 |
+
- `label` column renamed to `original_label`
|
11 |
+
- creation of the `label` column using the following mapping, using common practices ([1](https://github.com/rabeehk/robust-nli/blob/c32ff958d4df68ac2fad9bf990f70d30eab9f297/data/scripts/joci.py#L22-L27), [2](https://github.com/azpoliak/hypothesis-only-NLI/blob/b045230437b5ba74b9928ca2bac5e21ae57876b9/data/convert_joci.py#L7-L12))
|
12 |
+
|
13 |
+
```
|
14 |
+
{
|
15 |
+
0: "contradiction",
|
16 |
+
1: "contradiction",
|
17 |
+
2: "neutral",
|
18 |
+
3: "neutral",
|
19 |
+
4: "neutral",
|
20 |
+
5: "entailment",
|
21 |
+
}
|
22 |
+
```
|
23 |
+
|
24 |
+
- finally, converting this to the usual NLI classes, that is `{"entailment": 0, "neutral": 1, "contradiction": 2}`
|
25 |
+
|
26 |
+
|
27 |
+
## Code to create dataset
|
28 |
+
```python
|
29 |
+
import pandas as pd
|
30 |
+
from datasets import Features, Value, ClassLabel, Dataset
|
31 |
+
|
32 |
+
|
33 |
+
# read data
|
34 |
+
df = pd.read_csv("<path to folder>/joci.csv")
|
35 |
+
|
36 |
+
# column name to lower
|
37 |
+
df.columns = df.columns.str.lower()
|
38 |
+
|
39 |
+
# rename label column
|
40 |
+
df = df.rename(columns={"label": "original_label"})
|
41 |
+
|
42 |
+
# encode labels
|
43 |
+
df["label"] = df["original_label"].map({
|
44 |
+
0: "contradiction",
|
45 |
+
1: "contradiction",
|
46 |
+
2: "neutral",
|
47 |
+
3: "neutral",
|
48 |
+
4: "neutral",
|
49 |
+
5: "entailment",
|
50 |
+
})
|
51 |
+
|
52 |
+
# encode labels
|
53 |
+
df["label"] = df["label"].map({"entailment": 0, "neutral": 1, "contradiction": 2})
|
54 |
+
|
55 |
+
# cast to dataset
|
56 |
+
features = Features({
|
57 |
+
"context": Value(dtype="string"),
|
58 |
+
"hypothesis": Value(dtype="string"),
|
59 |
+
"label": ClassLabel(num_classes=3, names=["entailment", "neutral", "contradiction"]),
|
60 |
+
"original_label": Value(dtype="int32"),
|
61 |
+
"context_from": Value(dtype="string"),
|
62 |
+
"hypothesis_from": Value(dtype="string"),
|
63 |
+
"subset": Value(dtype="string"),
|
64 |
+
})
|
65 |
+
ds = Dataset.from_pandas(df, features=features)
|
66 |
+
ds.push_to_hub("joci", token="<token>")
|
67 |
+
```
|