mbayan commited on
Commit
f91ccb5
·
verified ·
1 Parent(s): 7e90876

Update dataset card

Browse files
Files changed (1) hide show
  1. README.md +13 -16
README.md CHANGED
@@ -26,7 +26,9 @@ configs:
26
  path: data/dev_test-*
27
  - split: test
28
  path: data/test-*
 
29
  dataset_info:
 
30
  features:
31
  - name: id
32
  dtype: string
@@ -40,19 +42,13 @@ dataset_info:
40
  sequence: string
41
  splits:
42
  - name: train
43
- num_bytes: 274812860.5
44
  num_examples: 3500
45
  - name: dev
46
- num_bytes: 40130069.0
47
  num_examples: 500
48
  - name: dev_test
49
- num_bytes: 38926730.0
50
  num_examples: 500
51
  - name: test
52
- num_bytes: 38669448.0
53
  num_examples: 500
54
- download_size: 386300009
55
- dataset_size: 392539107.5
56
  ---
57
 
58
  # ArGuard – Track A: Arabic Hateful Memes
@@ -85,12 +81,13 @@ Given a meme (image + Arabic text):
85
  |--------------|---------|--------------|-----------------------------------------|---------------------------------|
86
  | `train` | 3,500 | full | single-annotated bulk | development phase |
87
  | `dev` | 500 | full | single-annotated bulk | development phase |
88
- | `dev_test` | 500 | **dropped** | single-annotated test sample | development phase (leaderboard) |
89
  | `test` | 500 | **dropped** | **triple-annotated gold** (calibration) | final-evaluation phase (leaderboard) |
90
 
91
- - `dev_test` was the **leaderboard set** for the development phase, which
92
- is now closed. Labels are stripped (`label = null`,
93
- `fine_grained_label = []`).
 
94
  - `test` is the **blind test set** for the final-evaluation phase and is
95
  the file you submit predictions for. It is released the same way as
96
  `dev_test` — images and OCR text only, labels stripped. All 500 records
@@ -104,7 +101,7 @@ Given a meme (image + Arabic text):
104
  | train | 1,324 | 2,176 | 37.8% |
105
  | dev | 189 | 311 | 37.8% |
106
  | dev_test | 189 | 311 | 37.8% |
107
- | **Total** | **1,702** | **2,798** | 37.8% |
108
 
109
  The label distribution of the blind `test` split is withheld during the
110
  final-evaluation phase.
@@ -137,8 +134,8 @@ than the meme counts.
137
  "id": "f9a8…b1.jpg", # str – original image filename, unique
138
  "image": <PIL.Image.Image>, # embedded bytes, decoded on access
139
  "text": "…", # str – OCR-extracted Arabic meme text
140
- "label": "Hateful" | "Not Hateful" | None, # None on dev_test / test
141
- "fine_grained_label": [...], # list[str] – empty on dev_test / test
142
  }
143
  ```
144
 
@@ -154,13 +151,13 @@ train_ex = ds["train"][0]
154
  train_ex["image"].show()
155
  print(train_ex["text"], train_ex["label"], train_ex["fine_grained_label"])
156
 
157
- # dev_test and test are unlabelled — used only to produce submissions
158
  print(ds["test"][0]["label"]) # -> None
159
  ```
160
 
161
  The blind `test` split is the submission target for the final-evaluation
162
- phase. Train on `train` + `dev`, predict on `test`, and upload to
163
- CodaBench (links below).
164
 
165
  ## Shared-task resources
166
 
 
26
  path: data/dev_test-*
27
  - split: test
28
  path: data/test-*
29
+ default: true
30
  dataset_info:
31
+ config_name: default
32
  features:
33
  - name: id
34
  dtype: string
 
42
  sequence: string
43
  splits:
44
  - name: train
 
45
  num_examples: 3500
46
  - name: dev
 
47
  num_examples: 500
48
  - name: dev_test
 
49
  num_examples: 500
50
  - name: test
 
51
  num_examples: 500
 
 
52
  ---
53
 
54
  # ArGuard – Track A: Arabic Hateful Memes
 
81
  |--------------|---------|--------------|-----------------------------------------|---------------------------------|
82
  | `train` | 3,500 | full | single-annotated bulk | development phase |
83
  | `dev` | 500 | full | single-annotated bulk | development phase |
84
+ | `dev_test` | 500 | full | single-annotated test sample | development phase (leaderboard); **labels released** |
85
  | `test` | 500 | **dropped** | **triple-annotated gold** (calibration) | final-evaluation phase (leaderboard) |
86
 
87
+ - `dev_test` was the **leaderboard set** for the development phase. That
88
+ phase is now closed and its gold labels have been **released**, so
89
+ `dev_test` is usable as additional labelled data. It is disjoint from
90
+ `test`.
91
  - `test` is the **blind test set** for the final-evaluation phase and is
92
  the file you submit predictions for. It is released the same way as
93
  `dev_test` — images and OCR text only, labels stripped. All 500 records
 
101
  | train | 1,324 | 2,176 | 37.8% |
102
  | dev | 189 | 311 | 37.8% |
103
  | dev_test | 189 | 311 | 37.8% |
104
+ | **Total (labelled)** | **1,702** | **2,798** | 37.8% |
105
 
106
  The label distribution of the blind `test` split is withheld during the
107
  final-evaluation phase.
 
134
  "id": "f9a8…b1.jpg", # str – original image filename, unique
135
  "image": <PIL.Image.Image>, # embedded bytes, decoded on access
136
  "text": "…", # str – OCR-extracted Arabic meme text
137
+ "label": "Hateful" | "Not Hateful" | None, # None on the blind `test` split
138
+ "fine_grained_label": [...], # list[str] – empty on the blind `test` split
139
  }
140
  ```
141
 
 
151
  train_ex["image"].show()
152
  print(train_ex["text"], train_ex["label"], train_ex["fine_grained_label"])
153
 
154
+ # `test` is the only unlabelled split — used only to produce submissions
155
  print(ds["test"][0]["label"]) # -> None
156
  ```
157
 
158
  The blind `test` split is the submission target for the final-evaluation
159
+ phase. Train on `train` + `dev` + `dev_test`, predict on `test`, and
160
+ upload to CodaBench (links below).
161
 
162
  ## Shared-task resources
163